Infrared diagnostic method and system for power transmission and transformation equipment

Abstract

The application relates to the technical field of power equipment state monitoring and fault diagnosis, and discloses an infrared diagnosis method and system for power transmission and transformation equipment. The method comprises the following steps: collecting infrared thermal image data of the equipment and applying a filter to remove noise to obtain a clear image; adopting an image segmentation model to perform pixel-level segmentation, determining the boundaries of equipment components, extracting an initial feature vector, and containing texture and depth thermal distribution information; determining a temperature abnormal area based on the feature set, and performing material classification in combination with the texture information; analyzing the structural differences of the components through a classification model, obtaining structural feature descriptions, and further combining the structural features with the temperature abnormal area to perform fine feature matching, so that synchronous identification is realized, and the accuracy and real-time performance of fault diagnosis of the power transmission and transformation equipment are improved.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to an infrared diagnostic method and system for power transmission and transformation equipment. Background Technology

[0002] In modern society, with the continuous growth of energy demand, the normal operation of power transmission and transformation equipment has become a crucial factor in ensuring power supply. However, during long-term operation, power transmission and transformation equipment is often affected by various factors, leading to varying degrees of failure. These failures not only affect the operating efficiency of the equipment but can also, in severe cases, cause power outages, resulting in significant inconvenience to social production and people's lives. Therefore, how to monitor and diagnose the operating status of power transmission and transformation equipment in real time and accurately, and promptly identify and resolve potential faults, has become key to the safe operation of the power system.

[0003] Traditional methods for diagnosing faults in power transmission and transformation equipment mainly rely on manual inspections and periodic maintenance. These methods are not only labor-intensive and inefficient, but also struggle to detect potential problems in equipment operation in a timely manner. Furthermore, traditional methods have limitations in fault location and analysis, especially in complex operating environments and harsh working conditions, often failing to provide accurate diagnostic results.

[0004] In recent years, with the rapid development of infrared imaging technology, its application in equipment fault detection has gradually gained widespread attention. Infrared imaging technology can acquire the temperature distribution of equipment in real time, detecting hot spots and abnormal temperature changes, providing important evidence for equipment fault diagnosis. However, single infrared thermal image data often suffers from noise interference and lacks precise automated analysis methods. Therefore, how to effectively process and analyze infrared images to accurately determine the fault status of equipment has become a major challenge in infrared diagnostic technology.

[0005] Against this backdrop, artificial intelligence-based image processing and pattern recognition technologies have been widely applied in recent years. By combining deep learning algorithms such as convolutional neural networks, feature information in infrared images can be automatically extracted, enabling image segmentation, classification, and fault identification. However, these methods still face technical challenges in practical applications, such as how to accurately extract features of equipment components, how to handle complex noise interference, and how to perform multi-dimensional analysis.

[0006] Therefore, providing an infrared diagnostic method and system for power transmission and transformation equipment, which can overcome the shortcomings of existing technologies through multi-level and multi-scale feature extraction and classification analysis, and achieve rapid and accurate diagnosis of equipment faults in complex environments, is an urgent problem to be solved. Summary of the Invention

[0007] This application provides an infrared diagnostic method and system for power transmission and transformation equipment, which can improve the accuracy and real-time performance of fault diagnosis for power transmission and transformation equipment.

[0008] In a first aspect, this application provides an infrared diagnostic method for power transmission and transformation equipment, the method comprising:

[0009] The infrared thermal image data of the acquisition device is collected and a preset filter is applied to remove noise interference to obtain a preliminary clear infrared image;

[0010] Based on the preliminary clear infrared image, an image segmentation model is used to perform initial pixel-level segmentation to determine the boundary regions of device components in order to separate specific connection parts, and obtain the final segmented image;

[0011] An initial feature vector is extracted from the boundary region of the final segmented image. The initial feature vector includes texture information and depth heat distribution information to obtain a clustering feature set for labeling the closed mask.

[0012] If the original thermal distribution information in the feature set exceeds a preset temperature threshold, it is determined to be a temperature abnormality area, and the component material is further classified in combination with the texture information.

[0013] A classification model is used to analyze the structural differences in the materials of the classified components to obtain a description of their structural features.

[0014] By fusing the correspondence between the structural feature description and the temperature anomaly region, subtle feature matching is performed to obtain synchronous recognition results;

[0015] The parameters of the image segmentation model are updated based on the synchronous recognition results to obtain the optimized segmentation output, and fault diagnosis is performed based on the optimized segmentation output.

[0016] Secondly, this application provides an infrared diagnostic system for power transmission and transformation equipment, the system comprising:

[0017] The data acquisition module is used to acquire infrared thermal image data from the device and apply a preset filter to remove noise interference, thereby obtaining a preliminary clear infrared image.

[0018] The image segmentation module is used to perform initial pixel-level segmentation based on the preliminary clear infrared image using an image segmentation model, determine the boundary regions of device components to separate specific connection parts, and obtain the final segmented image; and extract initial feature vectors based on the boundary regions of the final segmented image, the initial feature vectors including texture information and depth thermal distribution information, to obtain a clustering feature set for labeling closed masks;

[0019] An anomaly analysis module is used to determine a temperature anomaly region if the original thermal distribution information in the feature set exceeds a preset temperature threshold, and further classify the component material in combination with the texture information; and to perform structural difference analysis on the classified component material using a classification model to obtain a structural feature description.

[0020] The feature fusion module is used to perform subtle feature matching by fusing the correspondence between the structural feature description and the temperature anomaly region to obtain synchronous recognition results;

[0021] The model optimization module is used to update the parameters of the image segmentation model according to the synchronous recognition result, obtain the optimized segmentation output, and perform fault diagnosis based on the optimized segmentation output.

[0022] The technical solution provided in this application employs a method combining infrared imaging technology and deep learning algorithms, which can effectively improve the accuracy and real-time performance of fault diagnosis for power transmission and transformation equipment. By performing noise removal, image segmentation, feature extraction, and classification analysis on infrared thermal image data, it can not only accurately identify abnormal temperature areas in the equipment but also precisely classify the differences in equipment materials and structures, thereby providing a comprehensive understanding of the equipment's operating status. Compared with traditional manual inspections and methods based on single infrared thermal image analysis, this technical solution has the following significant advantages: Automated image processing and feature extraction significantly reduce the need for manual intervention, improving diagnostic efficiency and accuracy; this solution can process high-noise infrared images in complex environments, ensuring the reliability of diagnostic results; the system can dynamically adjust the diagnostic model based on real-time data, further improving the accuracy of fault detection and the system's adaptability. Therefore, this technical solution can effectively improve the fault diagnosis capability of power transmission and transformation equipment, reduce the risk of outages caused by equipment failures, and provide strong support for the safe and stable operation of the power system. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of an infrared diagnostic method for power transmission and transformation equipment according to this application;

[0025] Figure 2 This is an example of an image acquired by an infrared diagnostic method for power transmission and transformation equipment according to this application;

[0026] Figure 3 This is a demonstration of the segmentation results of an infrared diagnostic method for power transmission and transformation equipment according to this application;

[0027] Figure 4 This is a flowchart of the component abnormality region synchronous identification method based on temperature gradient and structural feature matching in this application;

[0028] Figure 5 This is a schematic diagram of the structure of an infrared diagnostic system for power transmission and transformation equipment according to this application. Detailed Implementation

[0029] This application provides an infrared diagnostic method and system for power transmission and transformation equipment. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0030] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of an infrared diagnostic method for power transmission and transformation equipment in this application includes:

[0031] Step S1: Acquire infrared thermal image data from the device and apply a preset filter to remove noise interference to obtain a preliminary clear infrared image.

[0032] Specifically, after acquiring infrared thermal image data from the device, a preset filter is first applied to process the data to remove noise interference. The filter improves image quality by smoothing noise elements in the image, resulting in a preliminarily clear infrared image. This process includes suppressing unnecessary high-frequency noise in the image, making the temperature information of the device components more prominent, facilitating subsequent analysis and processing. After filtering, the resulting infrared image will have better contrast and clarity, providing a reliable data foundation for image segmentation and feature extraction.

[0033] Step S2: Based on the preliminary clear infrared image, use an image segmentation model to perform initial pixel-level segmentation, determine the boundary regions of device components to separate specific connection parts, and obtain the final segmented image.

[0034] Specifically, after obtaining a preliminary, clear infrared image, an image segmentation model is used to process the image. This model segments the image into multiple regions based on temperature changes and texture features. Through initial pixel-level segmentation, the system can accurately identify and delineate the boundary regions of device components, particularly separating specific areas at connection points. This process not only ensures clear boundaries for each device component but also effectively identifies various connection points within the device, thus providing accurate component information for analysis. After this segmentation process, the resulting image clearly marks the boundaries and connection points of each component within the device.

[0035] Step S3: Extract initial feature vectors based on the boundary regions of the final segmented image. The initial feature vectors include texture information and depth heat distribution information to obtain a clustering feature set for labeling the closed mask.

[0036] Specifically, based on the boundary regions of the final segmented image, the system first extracts texture information and depth thermal distribution information from the image. Texture information is obtained by analyzing the grayscale changes and structural features of different regions in the image, reflecting the detailed features of the device component surface; while depth thermal distribution information is obtained by analyzing the thermal distribution in the image to obtain the temperature change trends of different parts. This information is combined to form an initial feature vector, representing the thermal characteristics and surface structure of each device component. After obtaining the initial feature vector, the system processes the features using a clustering algorithm, grouping similar features to obtain a clustered feature set used to annotate the closed mask. These feature sets reflect the similarities and differences in temperature and texture among the various components of the device, providing crucial information for fault diagnosis.

[0037] Step S4: If the original thermal distribution information in the feature set exceeds the preset temperature threshold, it is determined to be a temperature abnormal area, and the component material is further classified in combination with the texture information.

[0038] Specifically, a feature set is acquired, and the raw thermal distribution information within it is analyzed. If the thermal distribution information of a certain area exceeds a preset temperature threshold, the system identifies that area as a temperature anomaly region. This process compares the temperature values ​​of various areas in the image with the preset threshold to determine whether there are components or areas with abnormally high temperatures. Once a temperature anomaly region is identified, it is analyzed in conjunction with texture information. Texture information reflects the physical characteristics of the equipment surface or components, such as the surface roughness and structural features of the material. By classifying and analyzing the texture information, the material type of the equipment corresponding to the temperature anomaly region can be determined. This process compares the texture features in the image with known component material information to accurately classify the component material, thereby providing more detailed information for fault diagnosis and helping to locate the cause of the fault.

[0039] Step S5: Use a classification model to analyze the structural differences in the materials of the classified components to obtain a description of the structural features.

[0040] Specifically, when analyzing the structural differences in the materials of categorized components, the system inputs the material characteristics of each categorized component into a pre-trained classification model. The model learns and analyzes information such as temperature distribution, texture features, and morphological characteristics of components made of different materials to identify structural differences between them. The classification model can extract differences in physical properties of materials, such as thermal expansion, thermal conductivity, and surface stress distribution. These differences are particularly pronounced in components made of different materials. For example, the difference in thermal expansion characteristics between metal and plastic components, or the strength changes of different alloy materials under high-temperature environments. The model further analyzes factors such as the geometry, surface texture, and roughness of the components, quantifying these structural differences and generating detailed structural feature descriptions. Furthermore, the analysis also considers factors such as material fatigue and deformation caused by thermal stress, thereby accurately revealing potential failure risks caused by material properties. These structural feature descriptions provide important evidence for fault diagnosis, effectively identifying faults caused by material mismatch or material aging, further improving the reliability and safety of equipment operation.

[0041] Step S6: By fusing structural feature descriptions and the correspondence between temperature anomaly areas, perform subtle feature matching to obtain synchronous recognition results.

[0042] Specifically, structural feature descriptions are matched with temperature anomaly regions to analyze the relationship between the two. Temperature anomaly regions are typically critical areas in equipment where failures may occur, while structural feature descriptions provide information about component materials, shapes, and thermal properties. By fusing these two elements, the system can further confirm whether these anomaly regions are related to the equipment's material, shape, or other physical properties based on known temperature anomaly regions and structural features. During the fusion process, the system employs subtle feature matching technology to accurately match minute differences between temperature anomaly regions and structural features. Subtle feature matching not only focuses on the overall trend of temperature distribution but also considers localized details such as local temperature differences, surface texture, and deformation. This matching helps the system accurately identify specific component features related to temperature anomalies, thereby more effectively locating potential fault sources. After feature matching, the system obtains synchronous identification results, accurately identifying structural problems related to temperature anomaly regions. Through this synchronous identification, the system can not only confirm which components have temperature anomalies but also reveal which specific structural features or material problems these anomalies may be related to, providing more detailed evidence for fault diagnosis and thus improving the accuracy and response speed of fault detection.

[0043] Step S7: Update the parameters of the image segmentation model according to the synchronous recognition results to obtain the optimized segmentation output, and perform fault diagnosis based on the optimized segmentation output.

[0044] Specifically, after obtaining the synchronous recognition results, the parameters of the image segmentation model are updated based on the relationship between the identified temperature anomaly regions and structural features. This process involves adjusting the model's segmentation rules and algorithm parameters according to the matching of identified structural features and temperature anomalies, in order to improve the model's sensitivity and recognition accuracy to fault regions. Through this update, the system can better cope with complex equipment state changes and identify more potential fault regions, especially in complex equipment operating environments or under multiple interference conditions. The updated image segmentation model, by optimizing its segmentation output, can generate more accurate segmented images, clearly marking information such as the boundaries of equipment components, connection points, and temperature anomaly regions. This optimized segmentation output can further help the system locate and confirm the faulty parts in the equipment and provide a more accurate basis for fault diagnosis. Based on the optimized segmentation output, the system can perform more detailed fault diagnosis. By analyzing the optimized image, combined with the equipment's operating status and known fault modes, the system can accurately identify the specific fault type of the equipment, locate the faulty component, and even predict potential fault risks. This fault diagnosis method based on optimized segmented output significantly improves the accuracy of fault detection, reduces the possibility of misdiagnosis, and can more effectively ensure the stable operation of power transmission and transformation equipment.

[0045] It is understood that the executing entity of this application can be an infrared diagnostic system for power transmission and transformation equipment, or it can be a terminal or a server; no specific limitation is made here. This application's embodiments use a server as an example for illustration.

[0046] In one specific embodiment, the process of performing step S1 may specifically include the following steps:

[0047] The raw data output by the infrared thermal imaging device is acquired, and the data deviation is corrected by the sensor calibration module to obtain the calibrated infrared data.

[0048] A mean filter is used to preprocess the calibrated infrared data to suppress high-frequency noise and generate smooth initial data.

[0049] If the signal quality of the initial data is lower than the preset quality threshold, it will be further processed by a median filter to obtain the denoised data.

[0050] Based on the denoised data, an edge detection algorithm is applied to extract the contour features of the infrared image and generate a feature-enhanced image.

[0051] By normalizing the grayscale to enhance the features of the image and adjusting the range of pixel values, a contrast-optimized image is obtained.

[0052] A convolutional neural network is used to perform pattern recognition on contrast-optimized images to determine whether there are abnormal regions and obtain classification results.

[0053] Based on the classification results, a preliminary, clear infrared image containing markers of anomalous regions is generated and output.

[0054] Specifically, the raw data output by the infrared thermal imaging equipment will be calibrated by a sensor calibration module to eliminate data deviations caused by sensor errors or environmental changes. (Reference) Figure 2 This image shows a sample of the acquired image. It assumes the use of an FLIRT series thermal imager, operating in the 7.5-13 micrometer band, with a resolution of 640x480 pixels, a frame rate of 30Hz, and a temperature measurement range of -20°C to 650°C. During acquisition, the ambient temperature compensation was set to 25°C and the relative humidity to 50% to correct for atmospheric absorption effects. The raw data is stored in 16-bit grayscale image format, with each pixel value corresponding to a temperature value ranging from 0 to 65535, linearly mapped to the temperature range of -20°C to 650°C. The calibrated data undergoes mean filtering. The mean filter uses a 3×3 pixel window, averaging the values ​​of the nine neighboring pixels of each pixel and replacing the value of the center pixel to eliminate random noise in the image. For example, if a pixel value is 45000 (approximately 300°C), and its neighboring pixel values ​​are [44500, 45100, 44900, 45000, 45200, 44800, 44700, 45300, 44600], the median of its 3×3 neighborhood is 45000, thus maintaining temperature consistency and reducing the impact of noise points. If the signal quality of the initial data is lower than a preset quality threshold, such as a signal-to-noise ratio lower than a set standard, a median filter is used for further noise reduction. The median filter replaces the center pixel value with the median of each pixel's neighborhood, effectively removing isolated noise points while preserving image edge details and reducing the blurring effect on edges.

[0055] On the denoised data, edge detection algorithms, such as the Sobel operator, are applied to extract the image's contour features. Edge detection significantly highlights the boundaries and contours of the device by detecting areas with large variations in image grayscale values. Image details are enhanced in this process, facilitating subsequent feature extraction and defect detection. Grayscale normalization is then applied to the feature-enhanced image, adjusting the pixel value range and expanding values ​​originally concentrated within a specific grayscale range to the entire grayscale range of 0 to 65535. This improves image contrast, making temperature changes of the device more apparent. The optimized image is then input into a convolutional neural network (CNN) for pattern recognition. Through training, the CNN learns to identify abnormal regions in the image and classifies them based on image features. For example, by learning different device materials and abnormal temperature patterns, the CNN can identify faulty areas in the device. Based on the classification results, the system determines whether the device is malfunctioning and outputs a preliminary, clear infrared image containing markers of abnormal regions. The image generated and output by the system not only includes the device's temperature distribution information but also clearly marks abnormal temperature regions for further fault diagnosis and repair.

[0056] In one specific embodiment, the process of performing step S2 may specifically include the following steps:

[0057] By using a pre-defined image segmentation model, the initial clear infrared image is subjected to boundary optimization processing, and the pixel distribution in the boundary region is adjusted to obtain an optimized segmented image.

[0058] A contour tracking algorithm is used to extract the boundaries of device components in the optimized segmented image in a refined manner, generating accurate boundary data;

[0059] Determine whether the continuity of the precise boundary data is lower than a preset continuity threshold. If so, smooth the precise boundary data using morphological processing to obtain a smooth boundary image.

[0060] Based on smooth boundary images, a region filling method is used to annotate specific connection parts with pixels to generate an annotated segmentation image;

[0061] Feature extraction of specific connection regions in the labeled segmented image is performed by grayscale analysis to obtain component feature data;

[0062] Determine whether the uniformity of the distribution of component feature data is lower than a preset distribution threshold. If so, group the feature data using a clustering algorithm to generate classified component data.

[0063] Based on the classified component data, an infrared image containing the location information of specific connection points is generated, resulting in the final segmented image.

[0064] Specifically, based on a preliminary, clear infrared image, a pre-defined image segmentation model is used to optimize the image's boundaries. By adjusting the pixel distribution in the boundary regions, the model optimizes the contours of the device components in the image, ensuring the accuracy and clarity of the segmentation. The optimized image is then used for refined extraction of the device component boundaries. This process is performed using a contour tracking algorithm to extract precise boundary data for the device components. The contour tracking algorithm effectively identifies and connects continuous edges in the image, avoiding boundary breaks and discontinuities. If the boundary data after refined extraction exhibits poor continuity, falling below a pre-defined continuity threshold, the system uses morphological processing to smooth the boundary data. Morphological processing removes noise from the image through erosion and dilation operations, ensuring smoother and more accurate boundaries for the device components.

[0065] Based on smooth boundary images, the system employs a region-filling method to annotate specific connection points with pixels, generating an annotated segmentation image. This method accurately identifies connection points, ensuring these critical areas are fully captured and marked. Gray-scale analysis is used to extract features from the connection points in the annotated segmentation image, obtaining thermal distribution and surface feature data for the components. This provides fundamental data for temperature analysis and fault diagnosis. For components with poor distribution uniformity, the system uses a clustering algorithm to group their feature data, further analyzing the differences between different components. The clustering algorithm effectively identifies and categorizes components with similar features, generating classified component data. Based on this classified component data, the system generates an infrared image containing the location information of specific connection points, resulting in the final segmentation image. This image not only accurately marks the location of components and connection points but also displays the temperature distribution, providing crucial information for fault diagnosis, overheat detection, and risk assessment. (Reference) Figure 3 The image shows the segmentation results, with the first image being the input image and the second image showing the segmentation results.

[0066] In one specific embodiment, the process of performing step S3 may specifically include the following steps:

[0067] Multi-scale feature extraction is performed on the boundary regions of the final segmented image using a convolutional neural network to obtain texture information and depth heat distribution information, and to generate an initial feature vector.

[0068] Principal component analysis is used to reduce the dimensionality of the initial eigenvectors, generating compressed eigenvectors;

[0069] Determine whether the variance of the compressed feature vector is lower than the preset variance threshold. If so, cluster the compressed feature vector using the mean shift algorithm to obtain a clustered feature set.

[0070] The support vector machine algorithm is used to classify the clustered feature set to determine the material type of the component;

[0071] The pixel distribution of the cluster feature set is analyzed by gray-level histogram analysis to generate a material distribution image;

[0072] Based on the material distribution image, a region growing algorithm is used to finely segment the material region to obtain the boundary point set of the fine material region.

[0073] A closed mask is generated based on the boundary point set of the fine material region. The pixels covered by the closed mask are assigned corresponding type labels according to the material type of the component, thus obtaining the final material classification image.

[0074] Specifically, the convolutional neural network performs multi-scale feature extraction on the boundary regions of the final segmented image to obtain texture information and depth thermal distribution information. Through convolution operations, local features in the image, such as edges, textures, shapes, and temperature variations, are extracted as feature vectors. Each convolutional layer uses a different size convolutional kernel, such as 3×3, to extract features at different scales and generate initial feature vectors. This vector represents the detailed features and temperature distribution of the device components. The feature extraction process is achieved through the following convolution formula:

[0075]

[0076] in, It is the k-th convolutional kernel. This represents the convolution operation. It is the first input image Layer feature map, It is the bias term of the k-th convolution kernel. It is the output feature vector.

[0077] Next, Principal Component Analysis (PCA) is used to reduce the dimensionality of the initial eigenvectors. PCA extracts the principal components with the largest variance by performing eigenvalue decomposition on the covariance matrix of the eigenvectors, thereby reducing the dimensionality of the data while retaining the most important information. Assume the initial eigenvectors are... The goal of PCA is to project these feature vectors into a new space to generate compressed feature vectors. The calculation formula is as follows:

[0078]

[0079] in, The eigenvector matrix represents the directions of the principal components. This is the dimensionality-reduced feature vector. In the dimensionality-reduced feature vector, the system determines whether its variance is below a preset threshold. If it is below the threshold, it indicates that the differences between features are not significant enough, and the system performs clustering using the Mean Shift algorithm. The Mean Shift algorithm finds pattern points in the data based on the density distribution of the feature vectors and groups similar features together. Specifically, the update formula for the Mean Shift algorithm is:

[0080]

[0081] in, For the first The current center of each cluster, It's a kernel function. These are data points in a cluster. These are the updated cluster centers. The feature set after clustering. The material type of the component is determined by classification using a Support Vector Machine (SVM). The SVM uses the following formula to classify the clustered feature set:

[0082]

[0083] in, For the weight vector, For the input feature vector, This is the bias term. SVM distinguishes the feature vectors of different material types by finding the optimal hyperplane, thus obtaining the material classification result. A gray-level histogram is used to analyze the pixel distribution of the clustered feature set, generating a material distribution image. The gray-level histogram reveals the distribution of material regions by calculating the pixel frequency at different gray levels. Finally, based on the material distribution image, a region growing algorithm is used to finely segment the material regions, obtaining a set of boundary points for the refined material regions. The basic idea of ​​the region growing algorithm is to start from a seed point and classify surrounding pixels based on similarity, using the following formula:

[0084]

[0085] in, For growth areas, It is a pixel grayscale value, It is a threshold. These are seed points. Based on the boundary point set of the fine-grained material region, a closed mask is generated, and pixels in the closed mask are assigned corresponding material labels according to the component's material type. This yields the final material classification image, where the material type of each component is clearly labeled, ensuring reliable data support for analysis and fault detection.

[0086] In one specific embodiment, the process of performing step S4 may specifically include the following steps:

[0087] The original thermal distribution information of the component is obtained by using an infrared imaging device to generate an initial thermal distribution image;

[0088] If the pixel values ​​in the initial thermal distribution image exceed the preset temperature threshold, they are marked as temperature anomaly areas, resulting in a set of anomaly areas.

[0089] Extract surface texture features from the set of abnormal regions to generate a texture feature vector;

[0090] The texture feature vectors are classified using a pre-trained convolutional neural network to obtain the material classification results;

[0091] Based on the material classification results, the mean shift algorithm is used to cluster the temperature anomaly areas to obtain the material type distribution;

[0092] Based on the material type distribution, a heat distribution image containing material type annotations is generated to obtain the final material distribution image;

[0093] The final material distribution image is segmented using a region growing algorithm to obtain a set of fine material regions.

[0094] Specifically, the thermal distribution information of the component is acquired using an infrared imaging device to generate an initial thermal distribution image. If the pixel values ​​in the image exceed a preset temperature threshold, they are marked as temperature anomaly areas, resulting in a set of anomaly areas. Surface texture features are extracted from these anomaly areas, and wavelet transform algorithms are used to extract the energy features of the low-frequency subband and the detail coefficients of the high-frequency subband, obtaining a texture feature vector that reflects the surface roughness and directionality of the component. The texture features are combined with the thermal distribution features to generate a feature vector containing both texture and thermal distribution. Dimensionality reduction is achieved through linear discriminant analysis to obtain a compressed feature vector. If the variance of the compressed feature vector is lower than a preset threshold, the mean-shift algorithm is used to cluster the feature vector, resulting in a clustered feature set. A support vector machine is used to classify the clustered feature set to determine the component material type, such as stainless steel, aluminum alloy, or titanium alloy.

[0095] Based on the material classification results, a heat distribution image with material type labels is generated. The image is then finely segmented using a region growing algorithm to obtain the boundaries of fine-grained material regions. Finally, a closed mask is generated based on these boundaries and labeled according to the component's material type, resulting in the final material classification image.

[0096] In one specific embodiment, the process of performing step S5 may specifically include the following steps:

[0097] An initial material image is generated by acquiring surface images of the component using an infrared imaging device.

[0098] Extract grayscale features from the initial material image to generate a grayscale feature set;

[0099] A pre-trained convolutional neural network is used to classify the gray-scale feature set to determine the material type of the component;

[0100] Determine whether the material type of the classified component is consistent with the preset standard material. If not, mark it as a material abnormal area and obtain a set of abnormal areas.

[0101] The mean shift algorithm is used to cluster the set of abnormal regions to obtain a material anomaly distribution map;

[0102] Based on the material anomaly distribution map, a region segmentation algorithm is used to generate fine material regions, and a set of segmented regions is obtained;

[0103] Specifically, a structural feature analysis of the segmented region set is performed using a support vector machine to obtain a structural feature description. Next, an infrared imaging device is used to acquire surface images of the component, generating an initial material image. This initial material image is analyzed to determine information such as temperature distribution and texture features on the component surface, providing a basis for subsequent material classification. Then, grayscale features are extracted from the initial material image to generate a grayscale feature set. These grayscale features reflect the changes in the component surface at different grayscale levels in the image, providing necessary data for texture analysis and material classification. A pre-trained convolutional neural network is used to classify the grayscale feature set to determine the component's material type. This process automatically extracts important features from the image through multi-layer processing of the convolutional neural network and determines the material type based on the trained model. If the classified component material type is inconsistent with the preset standard material, the system marks the component as a material anomaly region, obtaining an anomaly region set. This marking process helps the system identify and locate potential material anomalies, thereby attracting further attention. The system uses a mean-shift algorithm to cluster the anomaly region set to obtain a material anomaly distribution map. The mean-shift algorithm analyzes the distribution of anomalous regions in an image using density clustering, identifies the clustering patterns of anomalous regions of different materials, and generates a material anomaly distribution map. Based on this distribution map, a region segmentation algorithm is used to generate fine-grained material regions, resulting in a finely segmented set of regions.

[0104] Structural feature analysis of the segmented region set is performed using support vector machines to obtain a structural feature description. This analysis evaluates the structural differences of components by extracting features such as geometry, surface texture, and curvature. For example, for a mechanical component, its surface texture is analyzed using the gray-level co-occurrence matrix to obtain features such as contrast, correlation, and entropy. Furthermore, Canny edge detection is used to extract the component's edge contours and calculate the curvature distribution. These structural features are used to describe the component's geometric complexity and structural uniformity, ultimately generating a structural feature description that includes texture, geometry, and physical properties.

[0105] Through this series of processes, the system can complete the entire workflow from initial material image generation and abnormal area identification to material classification and structural feature analysis, providing accurate data support for subsequent fault diagnosis, fatigue life prediction, and component performance analysis. Taking a stainless steel gear as an example, the system accurately identifies its material type and analyzes its structural features through the above steps, thus obtaining the fatigue life prediction result for the component.

[0106] In one specific embodiment, the process of performing step S6 may specifically include the following steps:

[0107] An initial temperature image is generated by acquiring surface temperature images of the component using an infrared imaging device.

[0108] The initial temperature image is preprocessed using a mean filtering algorithm to obtain a smooth temperature image;

[0109] Extract temperature gradient features from smoothed temperature images to generate a set of temperature gradients;

[0110] Temperature gradient sets are classified using a pre-trained convolutional neural network to identify abnormal temperature regions;

[0111] If the correspondence between the temperature anomaly region and the structural feature description deviates from the preset temperature control threshold, it is marked as an abnormal matching region, and an abnormal matching set is obtained;

[0112] The region growing algorithm is used to segment the abnormal matching set to generate fine-grained matching regions;

[0113] By using support vector machines to perform feature analysis on the fine-matching region, synchronous recognition results are obtained.

[0114] Specifically, an initial temperature image is generated by acquiring surface temperature images of the component using an infrared imaging device. The image is then preprocessed using a mean filtering algorithm to remove noise, resulting in a smoothed temperature image. Temperature gradient features are extracted from the smoothed temperature image to generate a temperature gradient set, revealing the temperature variations on the component surface. For example, on the surface of a gear in a mechanical component, areas with large temperature gradients typically indicate potential overheating. A convolutional neural network is then used to classify the temperature gradient features to identify abnormal temperature areas; for instance, an abnormal temperature at a connection point of a gear might be related to overload or wear of the equipment.

[0115] If the correspondence between abnormal temperature regions and structural feature descriptions deviates from a preset temperature control threshold, the system marks these regions as abnormal matching regions and uses a region growing algorithm to finely segment them, obtaining finely matched regions. Structural feature analysis is then performed on the segmented finely matched regions using a support vector machine, ultimately outputting synchronous identification results. For example, a high-temperature region in a gear connection may be related to material fatigue or crack formation. By analyzing these abnormal matching regions, the system determines their relationship with the component's material and structure, providing a basis for thermal fatigue prediction and maintenance. (Reference) Figure 4 The figure illustrates a method for synchronous identification of abnormal regions in components based on temperature gradient and structural feature matching.

[0116] In one specific embodiment, the process of performing step S7 may specifically include the following steps:

[0117] A set of classification features is extracted from the synchronous recognition results, and a support vector machine is used to analyze the set of classification features to obtain the feature weight distribution.

[0118] The image segmentation model parameters are adjusted based on the feature weight distribution, and the adjusted parameters are verified using the random forest algorithm to generate a preliminary optimized model.

[0119] The segmentation boundary features are extracted from the preliminary optimization model, and the mean shift algorithm is used to cluster the segmentation boundary features to obtain the cluster boundary set;

[0120] Determine whether the matching degree between the cluster boundary set and the temperature distribution data is lower than the preset matching degree threshold. If so, adjust the segmentation boundary through an iterative optimization algorithm to generate an optimized segmentation output.

[0121] The abnormal region localization features are extracted from the optimized segmentation output, and the abnormal region localization features are classified by logistic regression to obtain a set of classified anomalies.

[0122] The diagnostic results are generated based on the set of anomalies. Data mapping technology is used to associate the diagnostic results with the fault diagnosis criteria to obtain the final diagnostic data.

[0123] Specifically, a set of classification features is extracted from the synchronous recognition results, and a support vector machine (SVM) is used to analyze these feature sets to obtain the feature weight distribution. The SVM learns the importance of different features to the classification results and determines the weight of each feature. Based on these feature weight distributions, the system adjusts the parameters of the image segmentation model to optimize the model for better identification of temperature anomaly regions. The adjusted parameters are validated using a random forest algorithm to generate a preliminary optimized model. The random forest constructs multiple decision trees and uses voting to evaluate the model's performance under different features, ensuring that the optimized parameters effectively improve segmentation accuracy.

[0124] The system extracts segmentation boundary features from the initial optimization model and clusters these features using the mean-shift algorithm to obtain a set of cluster boundaries. The mean-shift algorithm automatically clusters similar boundary points based on the density distribution of the features, thereby accurately identifying the equipment's segmentation boundaries. The system determines whether the matching degree between the cluster boundary set and the temperature distribution data is lower than a preset matching degree threshold. If the matching degree is low, it indicates that the current segmentation result may contain errors. At this point, the system adjusts the segmentation boundaries through an iterative optimization algorithm, generating an optimized segmentation output. Abnormal region location features are extracted from the optimized segmentation output, and logistic regression is used to classify these features, obtaining a set of classified anomalies. Based on the classified anomaly set, the system generates diagnostic results and associates these diagnostic results with fault diagnosis criteria through data mapping technology, thereby obtaining the final diagnostic data. This process ensures the accurate identification of abnormal temperature regions and the reliability of fault diagnosis results, providing a basis for equipment maintenance and fault prevention.

[0125] The above describes an infrared diagnostic method for power transmission and transformation equipment according to an embodiment of this application. The following describes an infrared diagnostic system for power transmission and transformation equipment according to an embodiment of this application. Please refer to [link / reference]. Figure 5 One embodiment of an infrared diagnostic system for power transmission and transformation equipment in this application includes:

[0126] The data acquisition module is used to acquire infrared thermal image data from the device and apply a preset filter to remove noise interference, thereby obtaining a preliminary clear infrared image.

[0127] The image segmentation module is used to perform initial pixel-level segmentation based on the preliminary clear infrared image using an image segmentation model, determine the boundary regions of device components to separate specific connection parts, and obtain the final segmented image; based on the boundary regions of the final segmented image, the module extracts initial feature vectors, which include texture information and depth thermal distribution information, to obtain a clustering feature set for labeling closed masks;

[0128] The anomaly analysis module is used to determine an abnormal temperature region if the original thermal distribution information in the feature set exceeds a preset temperature threshold, and further classify the component material by combining texture information; the classification model is used to perform structural difference analysis on the classified component material to obtain a structural feature description.

[0129] The feature fusion module is used to perform subtle feature matching by fusing structural feature descriptions with the correspondence between temperature anomaly regions and obtain synchronous recognition results;

[0130] The model optimization module is used to update the parameters of the image segmentation model based on the synchronous recognition results, obtain the optimized segmentation output, and perform fault diagnosis based on the optimized segmentation output.

[0131] Through the collaborative efforts of the aforementioned components, this embodiment enables efficient and accurate infrared diagnostics of power transmission and transformation equipment. The data acquisition module first acquires infrared thermal image data of the equipment and removes noise interference using a preset filter to obtain a preliminary, clear image, ensuring the quality of the foundational data for subsequent analysis. The image segmentation module performs pixel-level segmentation on the preliminary, clear infrared image, accurately determining the boundary regions of equipment components and extracting initial feature vectors based on these boundary regions. These feature vectors include texture information and depth thermal distribution information, providing a basis for anomaly detection and equipment classification. The anomaly analysis module, when processing the feature set, compares it with a preset temperature threshold to promptly identify temperature anomaly regions. It further analyzes the material of equipment components using texture information and classifies them, thus providing a physical basis for fault diagnosis. Structural difference analysis uses a classification model to deeply analyze the structural differences of different component materials, generating detailed structural feature descriptions. The feature fusion module fuses temperature anomaly regions and structural feature descriptions, obtaining synchronous recognition results through subtle feature matching, enhancing the system's ability to identify complex faults. The model optimization module dynamically adjusts the parameters of the image segmentation model based on the synchronous recognition results, enabling the system to self-optimize and improve segmentation accuracy and fault diagnosis accuracy. Through the efficient collaboration of its modules, the overall system provides an automated and accurate infrared diagnostic method for power transmission and transformation equipment, which can effectively improve the real-time performance and accuracy of fault detection and reduce the potential risks of equipment failures to the power system.

[0132] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An infrared diagnosis method for power transmission and transformation equipment, characterized in that, The method comprises: Step S1, collecting infrared thermal image data of the equipment and applying a preset filter to remove noise interference to obtain a preliminary clear infrared image; Step S2, performing initial pixel-level segmentation on the preliminary clear infrared image according to an image segmentation model to determine the boundary region of the equipment component to separate the specific connection part, and obtaining a final segmentation image; Step S3, extracting an initial feature vector based on the boundary region of the final segmentation image, wherein the initial feature vector comprises texture information and depth thermal distribution information, and obtaining a clustering feature set for labeling a closed mask; Step S4, if the original thermal distribution information in the feature set exceeds a preset temperature threshold, determining a temperature abnormal region, and further classifying the component material in combination with the texture information; Step S5, performing structure difference analysis on the classified component material by using a classification model to obtain a structure feature description; Step S6, performing fine feature matching by fusing the corresponding relationship between the structure feature description and the temperature abnormal region to obtain a synchronous recognition result; Step S7, updating the parameters of the image segmentation model according to the synchronous recognition result to obtain an optimized segmentation output, and performing fault diagnosis based on the optimized segmentation output.

2. The method according to claim 1, wherein The S1 comprises: Obtaining original data output by an infrared thermal imaging equipment, correcting data deviation through a sensor calibration module to obtain calibrated infrared data; Using a mean filter to pre-process the calibrated infrared data to suppress high-frequency noise and generate smooth initial data; If the signal quality of the initial data is lower than a preset quality threshold, further processing is performed through a median filter to obtain denoised data; According to the denoised data, an edge detection algorithm is applied to extract the contour features of the infrared image to generate a feature-enhanced image; The feature-enhanced image is processed by gray scale normalization to adjust the pixel value range to obtain a contrast-optimized image; A convolutional neural network is used to perform pattern recognition on the contrast-optimized image to determine whether there is an abnormal region and obtain a classification result; According to the classification result, the preliminary clear infrared image containing the abnormal region label is generated and output.

3. The method of claim 1, wherein the infrared diagnosis method is used for power transmission and transformation equipment. The S2 comprises: Optimizing the boundary of the preliminary clear infrared image through a preset image segmentation model to adjust the pixel distribution of the boundary region and obtain an optimized segmentation image; Using a contour tracking algorithm to finely extract the boundary of the equipment component in the optimized segmentation image to generate accurate boundary data; Determining whether the continuity of the accurate boundary data is lower than a preset continuity threshold, if yes, smoothing the accurate boundary data through a morphological processing method to obtain a smooth boundary image; Based on the smooth boundary image, using a region filling method to perform pixel labeling on the specific connection part to generate a labeled segmentation image; Extracting features of the specific connection part in the labeled segmentation image through gray scale analysis to obtain component feature data; Determining whether the distribution uniformity of the component feature data is lower than a preset distribution threshold, if yes, grouping the feature data through a clustering algorithm to generate classified component data; According to the classification component data, an infrared image containing position information of the specific connection site is generated, and a final segmentation image is obtained.

4. The method of claim 3, wherein the infrared diagnosis method is characterized by, The S3 comprises: Multi-scale feature extraction is performed on a boundary region of the final segmentation image by a convolutional neural network to obtain texture information and deep heat distribution information, and an initial feature vector is generated; Principal component analysis is used to perform dimension reduction processing on the initial feature vector to generate a compressed feature vector; It is determined whether a variance of the compressed feature vector is lower than a preset variance threshold, and if so, the compressed feature vector is clustered by a mean shift algorithm to obtain a clustered feature set; A support vector machine algorithm is used to classify the clustered feature set to determine a component material type; A material distribution image is generated by analyzing pixel distribution of the clustered feature set through a gray histogram; According to the material distribution image, a region growing algorithm is used to finely segment a material region to obtain a boundary point set of the fine material region; The closed mask is generated based on the boundary point set of the fine material region, and pixels covered by the closed mask are assigned corresponding type labels according to the component material type, so that a final material classification image is obtained.

5. The method of claim 1, wherein the power transmission and distribution equipment is selected from the group consisting of transformers, circuit breakers, and power lines. The S4 comprises: Original heat distribution information of a component is obtained by an infrared imaging device to generate an initial heat distribution image; If a pixel value in the initial heat distribution image exceeds the preset temperature threshold, the pixel value is marked as a temperature abnormal region to obtain an abnormal region set; Surface texture features are extracted from the abnormal region set to generate a texture feature vector; The texture feature vector is classified by a pre-trained convolutional neural network to obtain a material classification result; According to the material classification result, the mean shift algorithm is used to cluster the temperature abnormal region to obtain a material type distribution; Based on the material type distribution, a heat distribution image containing a material type label is generated to obtain a final material distribution image; A region growing algorithm is used to segment the final material distribution image to obtain a fine material region set.

6. The method of claim 1, wherein the power transmission and distribution equipment is selected from the group consisting of transformers, circuit breakers, and power lines. The S5 comprises: A component surface image is obtained by an infrared imaging device to generate an initial material image; Gray value features are extracted from the initial material image to generate a gray feature set; A pre-trained convolutional neural network is used to classify the gray feature set to determine the component material type; It is determined whether the classified component material type is consistent with a preset standard material, and if not, the component material type is marked as a material abnormal region to obtain an abnormal region set; A mean shift algorithm is used to cluster the abnormal region set to obtain a material abnormal distribution map; According to the material abnormal distribution map, a region segmentation algorithm is used to generate a fine material region to obtain a segmentation region set; A support vector machine is used to analyze structural features of the segmentation region set to obtain structural feature descriptions.

7. The method of claim 6, wherein the infrared diagnosis method is characterized by, The S6 comprises: A component surface temperature image is obtained by an infrared imaging device to generate an initial temperature image; A mean filter algorithm is used to pre-process the initial temperature image to obtain a smoothed temperature image; Temperature gradient features are extracted from the smoothed temperature image to generate a temperature gradient set; The temperature gradient set is classified by a pre-trained convolutional neural network to determine a temperature anomaly region; If the corresponding relationship between the temperature anomaly region and the structural feature description deviates from a preset temperature control threshold, the temperature anomaly region is marked as an abnormal matching region to obtain an abnormal matching set; The abnormal matching set is segmented by a region growing algorithm to generate a fine matching region; The fine matching region is analyzed by a support vector machine to obtain the synchronous recognition result.

8. The method of claim 1, wherein the power transmission and distribution equipment is selected from the group consisting of transformers, circuit breakers, and power lines. The S7 includes: Classification features are extracted from the synchronous recognition result, and the classification features are analyzed by a support vector machine to obtain a feature weight distribution; The image segmentation model parameters are adjusted according to the feature weight distribution, and the adjusted parameters are verified by a random forest algorithm to generate a preliminary optimized model; Segmentation boundary features are extracted from the preliminary optimized model, and the segmentation boundary features are clustered by a mean shift algorithm to obtain a clustered boundary set; It is determined whether the matching degree of the clustered boundary set and the temperature distribution data is lower than a preset matching degree threshold, and if so, the segmentation boundary is adjusted by an iterative optimization algorithm to generate the optimized segmentation output; Abnormal region positioning features are extracted from the optimized segmentation output, and the abnormal region positioning features are classified by a logistic regression to obtain a classified abnormal set; A diagnosis result output is generated according to the classified abnormal set, and the diagnosis result output is associated with a fault diagnosis basis by a data mapping technology to obtain final diagnosis data.

9. An infrared diagnosis system for power transmission and transformation equipment, which is used to realize the method for infrared diagnosis of power transmission and transformation equipment according to any one of claims 1-8, characterized in that, The infrared diagnosis system for power transmission and transformation equipment includes: A data acquisition module is configured to acquire equipment infrared thermal image data and apply a preset filter to remove noise interference to obtain a preliminary clear infrared image; An image segmentation module is configured to perform initial pixel-level segmentation on the preliminary clear infrared image using an image segmentation model to determine the boundary region of the equipment component to separate the specific connection part and obtain a final segmentation image; an initial feature vector is extracted based on the boundary region of the final segmentation image, the initial feature vector includes texture information and depth thermal distribution information, and a clustering feature set for labeling a closed mask is obtained; An abnormality analysis module is configured to determine a temperature anomaly region if the original thermal distribution information in the feature set exceeds a preset temperature threshold, and further classify the component material in combination with the texture information; a classification model is used to analyze the structural difference of the classified component material to obtain a structural feature description; A feature fusion module is configured to perform fine feature matching by fusing the corresponding relationship between the structural feature description and the temperature anomaly region to obtain a synchronous recognition result; A model optimization module is configured to update the parameters of the image segmentation model according to the synchronous recognition result to obtain an optimized segmentation output, and perform fault diagnosis based on the optimized segmentation output.

Images