Insulator aging state identification method and system based on comparative learning

Abstract

The invention relates to the technical field of power equipment operation and maintenance, in particular to an insulator aging state identification method and system based on comparative learning. The method introduces a comparative learning mechanism, and improves the accuracy and robustness of aging state recognition through the comparative analysis of the historical data and the current state. According to the invention, through multi-dimensional defect identification and quantitative scoring, special scoring is carried out from four dimensions of surface damage, electric erosion, metal corrosion and structure abnormity, a comprehensive abnormity scoring model is constructed, refined quantification of the aging state is realized, and the identification precision is improved. According to the method, the unique identity label of the insulator is bound with historical data through the two-dimensional code / OCR, and state evolution tracking and trend analysis can be performed on the same insulator, so that the aging state and the aging reason of the insulator are deeply analyzed, the service life of the insulator is predicted, and an information basis is provided for maintenance and inspection planning of the insulator.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment operation and maintenance, in particular to an insulator aging state recognition method and system based on contrast learning. BACKGROUND

[0002] Insulators are key components in power transmission lines for supporting conductors and preventing current from returning to the ground. Their operating state directly affects the safety and stability of the power system. As the operating life increases, insulators are exposed to natural environments for a long time and are affected by factors such as electric field, mechanical stress, ultraviolet light, acid rain, contamination, and temperature difference, which can easily cause aging phenomena such as umbrella disc cracking, electric erosion, metal accessory corrosion, and structural deformation. If these defects are not discovered and addressed in a timely manner, they may cause insulation failure, flashover, string breakage, and even large-scale power outages.

[0003] Currently, the detection of insulator aging state mainly relies on manual inspection, infrared temperature measurement, ultraviolet imaging, and ultrasonic detection. However, these methods have the following shortcomings: 1. Manual inspection is low in efficiency and high in cost, and is greatly influenced by terrain, weather, and personnel experience, making it difficult to achieve large-scale, high-frequency, and accurate detection; 2. Traditional image recognition methods have poor generalization ability and their recognition accuracy decreases under conditions of complex background, light variation, and perspective shift; 3. Existing detection methods lack unified quantitative evaluation standards, making it impossible to finely grade and predict the aging degree of insulators; 4. Existing detection methods do not establish an identity tracking mechanism, making it difficult to achieve individual-level historical data tracing and state evolution analysis.

[0004] Therefore, there is an urgent need for an automatic, high-precision, traceable, and predictable insulator aging state recognition method to improve the intelligent level and safety of power grid operation and maintenance. SUMMARY

[0005] The present application aims to provide an insulator aging state recognition method and system based on contrast learning, which solves the problems of low recognition accuracy, non-uniform evaluation standards, lack of trend prediction and individual tracking ability in the prior art, and realizes high-precision recognition, quantitative scoring, trend prediction, and maintenance decision support of insulator aging state.

[0006] To solve the above technical problems, the present application adopts the following technical solutions: An insulator aging state recognition method based on contrast learning, comprising the following steps: S1. Collecting high-definition images of insulators; S2. Identifying insulators from the images and assigning them unique identity labels; S3. Preprocessing the images, extracting the umbrella disc contour, and performing perspective transformation to obtain orthographic view images; S4. Defect recognition is performed on the pre-processed image, including surface damage recognition, electric corrosion damage recognition, metal accessory corrosion recognition, and overall structure abnormality recognition, and corresponding special scores are generated respectively; S5. The special scores are weighted and fused to obtain a comprehensive abnormality score of the insulator; S6. Based on the comprehensive abnormality score and historical data, the health status of the insulator is evaluated, the change trend is analyzed, and early warning and maintenance suggestions are generated.

[0007] Preferably, in step S2, the insulator and its accompanying weather-resistant two-dimensional code or digital coding card are photographed by a drone, ZBar or PaddleOCR is used to recognize the coding information, and the central database is matched or a new file is created.

[0008] Preferably, in step S3, Mask R-CNN is used for instance segmentation, the umbrella disc contour is extracted, and the image deformation is corrected by ellipse fitting and perspective transformation to obtain a front view angle.

[0009] Preferably, in step S4: the surface damage recognition uses U-Net to segment the damaged area, and calculates the surface damage score based on the damaged area and position; the electric corrosion damage recognition calculates the score of the electric corrosion damage and carbonization area according to the insulator material; the metal accessory corrosion recognition uses YOLOv8 to locate the metal parts, and calculates the corrosion score based on the spot, glossiness and integrity; the overall structure abnormality recognition uses YOLOv8 for axis fitting, curvature calculation and foreign object detection to generate a structure abnormality score.

[0010] Preferably, in step S5, the special scores include: the surface damage score , the electric corrosion damage score , the metal accessory corrosion score , and the overall structure abnormality score are weighted and fused to obtain a comprehensive abnormality score of the insulator The calculation method of the comprehensive abnormality score is: , wherein, are the weight coefficients of surface damage, electric corrosion damage, metal accessory corrosion and overall structure abnormality, respectively, .

[0011] Preferably, step S6 includes: dividing the health level according to the comprehensive abnormality score, and generating early warning and maintenance suggestions; calling historical data for time series analysis, calculating the trend coefficient, and identifying the dominant deterioration factor; based on the current score and trend prediction, the maintenance time window is determined.

[0012] Preferably, the method for calculating the trend coefficient in the S6 step is to select the last N times of inspection data, take time as the independent variable, and P value as the dependent variable to perform linear regression, and the slope k is the trend coefficient; the calculation method of the trend coefficient is: , is the time point of the first inspection, is the time average, is the abnormal score corresponding to the inspection time point, is the average of the abnormal score.

[0013] Preferably, the method for predicting the maintenance time window in the S6 step is to generate a change curve of the comprehensive abnormal score P based on the current comprehensive abnormal score P and the corresponding change trend coefficient k, and to predict the time limit for maintenance, and the calculation method is: , In the formula, is the predicted time, is the change trend of the comprehensive abnormal score P at the current time point, is the value of the comprehensive abnormal score P at the current time point, is the change rate of the k value, which is calculated based on the last N times of inspection data: , In the formula, is the time point of the first inspection, is the time average, is the k value at the first inspection, is the average of the k value.

[0014] The application also discloses an insulator aging state recognition system based on contrast learning, comprising: an image acquisition module for acquiring high-definition images of insulators; an identity recognition and data transmission module for recognizing the identity of the insulators and transmitting data; an image preprocessing module for image segmentation and perspective correction; a defect recognition and analysis module for recognizing various defects and generating special scores; a data fusion and score calculation module for calculating comprehensive abnormal scores and managing historical data; a state evaluation and early warning output module for health evaluation, trend analysis and report generation.

[0015] The application has the following advantages: 1. The application introduces a contrast learning mechanism, which improves the accuracy and robustness of aging state recognition through contrast analysis of historical data and current state. The application quantifies the aging state in detail and improves the recognition accuracy by constructing a comprehensive abnormal score model through multi-dimensional defect recognition and quantitative scoring in four dimensions of surface damage, electric corrosion, metal corrosion and structural abnormality. 2. The application realizes the binding of the unique identity of the insulator and the historical data through the two-dimensional code / OCR, can track the state evolution and trend analysis of the same insulator, thereby deeply analyzes the aging state and aging reason of the insulator, and predicts the service life of the insulator, provides information basis for the maintenance and inspection planning of the insulator. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the insulator aging state recognition method based on contrast learning of the application.

[0017] Figure 2 The system flowchart of the application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme of the application will be described clearly and completely in combination with the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, not all the embodiments. The components of the embodiments of the application described and shown in the drawings can be arranged and designed in various different configurations.

[0019] An insulator aging state recognition method based on contrast learning, comprising the following steps: S1. Collecting high-definition images of insulators.

[0020] A high-performance optical camera is used to collect multi-angle, high-resolution images of insulators in the power transmission line. The unmanned aerial vehicle is used to take pictures of the insulator from three angles, ensuring that the surface features of the key parts such as the insulator umbrella skirt, steel foot and ball head are covered, and ensuring that the complete image information of the insulator 360° view is captured, avoiding the appearance of blind area.

[0021] The image shooting method can be adjusted according to the length of the insulator. For short insulators, such as insulators with 3-4 umbrella plates, the insulator image can be collected by hovering on one side of the insulator. For long insulators, such as insulators with more than 5 umbrella plates, a single hovering point cannot capture the complete insulator image. The method of moving along the axis of the insulator and continuously shooting is used to ensure that each section of the umbrella skirt is captured completely.

[0022] After shooting, a preliminary check is performed to identify any issues such as severe blurring, overexposure, or underexposure in the images. Any problematic targets are then re-captured. The acquired raw image data, along with its metadata, is transmitted to the database.

[0023] S2. Identify insulators from the acquired high-definition images and assign or associate them with a unique identifier.

[0024] Each insulator is equipped with a weather-resistant QR code or digital code tag, which is installed on the crossarm or hardware of the tower near the insulator. The QR code or code tag is photographed simultaneously when a drone is used to collect high-resolution images of the insulator.

[0025] Before image processing, the encoded information on the insulator is read using a QR code recognition algorithm (ZBar) or an optical character recognition model (PaddleOCR). This encoded information serves as the unique ID of the insulator. This ID is then matched against a central database. If a record for that ID exists in the database, a direct association is established; otherwise, if it is a new ID (e.g., for a newly installed insulator), a new record is created in the database.

[0026] S3. Image Preprocessing Because there is an angular discrepancy between the drone's image capture and the parachute, the captured images need to be preprocessed to extract features and perform perspective transformation to correct the deformation caused by the tilted viewpoint. The specific method is as follows: Images of insulators with the same ID are input into the instance segmentation model Mask R-CNN. Mask R-CNN segments the images, generates a mask for the umbrella disk, and extracts the contour information of the umbrella disk from the mask.

[0027] The outermost contour is extracted from the binary mask using the `findContours` function in OpenCV. The contour extracted by `findContours` consists of a series of pixel coordinates. At least 5 pixels are randomly selected from these pixels. Then, the `cv2.fitEllipse` function in OpenCV is used to fit the selected pixels into an ellipse. Finally, a perspective transformation is performed based on the ellipse to map it into a circle, thus obtaining the frontal view of the image.

[0028] When selecting pixels to fit an ellipse, the selection process is repeated multiple times, each time choosing a different combination of pixels to improve the stability and accuracy of the fitting. Finally, the average of the multiple fitting results is taken as the optimal ellipse parameters, ensuring that the umbrella disk contour maintains geometric consistency after transformation.

[0029] S4. Image Analysis and Defect Recognition S4-1. Surface Damage Identification The pre-processed insulator front view image is analyzed using a U-Net-based semantic segmentation network to generate a binary mask. The white pixel area (value 255) is identified as a damaged area, and the black pixel area (value 0) is identified as a normal area.

[0030] The bounding rectangle of the damaged area is calculated, and the area of the bounding rectangle is calculated. The area ratio of the bounding rectangle in the entire image is set to exist threshold a, which can be set by the user, and the default value is 0.1%. When the area ratio is less than a, it is determined that there is no significant damage on the surface of the insulator; when the area ratio is greater than or equal to a, it is determined that there is a damage defect, and the damage position and the area of the bounding rectangle of the damage are recorded, and a surface damage score is generated according to the damage position and the area of the bounding rectangle of the damage , the calculation method is as follows: , In the formula, is the influence of the damage area on the surface damage score, is the weight of the area factor in the surface damage score, are the width and height of the bounding rectangle of the damaged area, respectively, is the area of the mask after mapping processing; is the influence of the damage position on the surface damage score, is the weight of the distance factor in the surface damage score, and are the horizontal and vertical coordinates of the center point of the bounding rectangle of the damaged area, respectively, and are the horizontal and vertical coordinates of the center of the umbrella disc after mapping processing, respectively, is the radius of the umbrella disc. and can be set by the user or determined based on machine learning.

[0031] The present scheme uses a bounding rectangle instead of the actual area of the damage. Although the value will be larger when calculating the area of the umbrella disc gap, it can fully reflect the complete coverage area of the umbrella disc crack. The umbrella disc crack itself has a small area, but the crack covers a wide area, which affects the strength of the umbrella disc in a large range. Using the bounding rectangle to calculate the crack coverage area can provide more adequate early warning of potential structural risks and avoid underestimating the impact of cracks on the overall mechanical properties of the umbrella disc by only focusing on the actual pixel area of the cracks. At the same time, the bounding rectangle calculation method does not require complex pixel-level accurate fitting of irregular crack edges, significantly reduces the computational complexity of image processing, and improves the real-time performance of surface damage recognition under the premise of ensuring the effectiveness of risk identification.

[0032] S4-2. Electric erosion damage recognition The preprocessed front view image of the insulator was analyzed using a U-Net-based semantic segmentation network to segment out the electro-erosion pulverization area or carbonization trace area.

[0033] When ceramic insulators experience electro-erosion, they typically generate electro-erosion chalking areas. First, the area of ​​the electro-erosion chalking area is calculated. Then, the color of the chalking area is analyzed by calculating the average saturation and brightness of its pixels in the HSV color space and comparing it with the base area of ​​the umbrella disk to calculate the degree of whitening. Finally, the texture features of the electro-erosion chalking area are calculated and compared with the smooth surface of the umbrella disk's base area to calculate the degree of roughness. Finally, an electro-erosion score is generated based on the area, whitening degree, and roughness of the electro-erosion chalking area. The specific calculation method is as follows: , In the formula, The impact of the area of ​​the pulverized region caused by electrical erosion on the electrical erosion identification score. The weights of the area factor for the electro-erosion pulverized region. The area of ​​the electrolytic erosion pulverization zone, The area of ​​the mask after mapping is denoted as .

[0034] To illustrate the impact of whitening degree on the electrical erosion identification score, The weights of the whitening degree factor, and These represent the average saturation and average brightness of the electro-erosion pulverized region, respectively. and These represent the average saturation and average brightness of the umbrella-shaped base region, respectively.

[0035] The influence of surface roughness on the electrical erosion identification score. The weight of the surface roughness factor in the electro-erosion pulverized zone. The contrast value of the electro-erosion pulverized region. This represents the contrast value of the umbrella disc base area.

[0036] For ceramic insulators, this solution uses three dimensions—area, color, and texture—to perform a quantitative and automated assessment of the insulator's electrical erosion damage. Finally, these three factors are integrated to generate a comprehensive electrical erosion damage score, thereby achieving an accurate determination of the insulator's electrical erosion condition. Where, , , It can be set manually or determined based on machine learning.

[0037] If the insulator is made of composite materials, its electrical erosion damage is usually manifested as carbonization, and its electrical erosion damage score is... This can be calculated using the carbonization area: , In the formula, The weights of the area factor of the carbonized region in the electro-erosion damage are... The area of ​​the electrolytic erosion pulverization zone, The area of ​​the mask after mapping is denoted as .

[0038] S4-3. Corrosion Identification of Metal Accessories Corrosion of metal components on insulators includes chemical corrosion and electrolytic corrosion. Chemical corrosion is caused by the action of air and moisture, resulting in pitting, mottled appearance, and loss of original luster. Electrolytic corrosion is caused by electrochemical reactions or the heat energy from electrical energy conversion, manifesting as localized ablation spots, pits, edge melting, or metal particle splashing and loss. In severe cases, perforation and fracture may occur, leaving traces of electrolytic corrosion on the surface. Both types of corrosion can cause spots, reduced luster, and damage to the surface integrity of metal components.

[0039] The YOLOv8 target detection model was used to accurately locate the bounding boxes of the steel caps and feet on the insulators in the image. The spots, gloss changes and integrity of the metal parts were detected, and a corrosion score of the metal accessories was generated. The calculation formula is as follows: , in, The influence weight of spotting factors on the corrosion score of metal accessories is given. Scoring of spots is used to quantify pitting, mottled, and ablation spots and pits in chemical corrosion and electrical erosion. , The pixel area of ​​the eroded region. This represents the total pixel area of ​​the metal component surface.

[0040] The weighting of the effect of gloss change on the corrosion score of metal accessories. A gloss variation score is given to quantify the gloss variation caused by a uniform chemical etching layer (such as oxides or hydroxides). , The average brightness of the current component area. For reference standard brightness.

[0041] The weighting of the influence of the integrity of metal parts on the corrosion score of metal accessories. The integrity dimension score is used to quantify structural damage such as edge melting, perforation, and fracture caused by electrical erosion.

[0042] , and These are the indicators for the presence of perforation and the presence of edge melting, both of which are binary variables. A value of 1 indicates the presence of either perforation or edge melting; a value of 0 indicates the absence of either. The missing rate of edge materials, Let be the theoretical perimeter of the metal accessory. The actual perimeter of the contour obtained from image segmentation. , , These are coefficients, which can be set manually or determined based on machine learning.

[0043] S4-4. Identification of Overall Structural Anomalies YOLOv8 was used to perform centerline fitting and tilt angle calculation, insulator disc bending degree calculation, and foreign object calculation on the insulator image, and an overall structural anomaly score was generated based on the calculation results. The calculation method is as follows: , in, The weighting of the tilt factor on the overall structural anomaly score. For tilt abnormality scoring, The inclination angle of the insulator's central axis. This is the reference suspension tilt angle of the insulator at the corresponding position. This is the threshold for abnormal tilt angles. When the actual tilt angle... With reference tilt angle The absolute difference exceeds the threshold At that time, it was considered that there was an abnormal tilt.

[0044] The weighting of the curvature factor on the overall structural anomaly score. The average curvature of the insulator is represented by . YOLOv8 first extracts the mask skeleton of the insulator, uses the least squares method to fit the skeleton to a straight line, and then calculates the average distance from each point on the skeleton line to the fitted straight line. For the first on the skeleton line The perpendicular distance from each point to the fitted line. This represents the total number of points on the skeleton line.

[0045] The weighting of foreign object factors on the overall structural anomaly score. The foreign object anomaly score is calculated by using YOLOv8 to detect the number of foreign objects on the insulator surface and surrounding area, and the percentage of the area covered. Weighting coefficients are used for each factor. , , It can be set manually or determined based on machine learning.

[0046] S5. Data Fusion Surface damage rating Electrolytic erosion score Corrosion rating of metal accessories and overall structural anomaly score Weighted fusion is performed to obtain the comprehensive anomaly score of the insulator. The calculation method is as follows: , In the formula, These are the weighting coefficients for surface damage, electrolytic corrosion, metal accessory corrosion, and overall structural abnormalities, respectively. Under normal circumstances, The value range is 0.2-0.3. The value range is 0.2-0.3. The value range is 0.3-0.4. The value range is 0.3-0.4. The above weighting coefficients can be dynamically adjusted according to the actual inspection environment and insulator type, or they can be optimized and determined based on historical inspection data and fault samples through machine learning models.

[0047] Insulator comprehensive anomaly score and the corresponding surface damage score Electrolytic erosion score Corrosion rating of metal accessories and overall structural anomaly score The data is transferred to the historical archive database for subsequent inspection analysis and trend prediction.

[0048] S6. Insulator Condition Assessment and Early Warning Based on the current comprehensive anomaly score of insulators Assess the health status of insulators. Simultaneously, retrieve historical data from the database for longitudinal comparison, analyze trends in anomaly scores, and generate a comprehensive diagnostic report including health scores and predictive maintenance recommendations.

[0049] S6-1. Insulator Current Health Status Assessment The comprehensive anomaly score of insulators calculated based on the current inspection cycle. By setting threshold ranges, a scale is established to classify the state of insulators.

[0050] Table 1. Insulator Condition Health Level Classification Standard

[0051] Based on the current comprehensive anomaly score Based on the location of the insulator, determine its health status level (HL) and generate corresponding warning levels and maintenance recommendations.

[0052] S6-2. Vertical Trend Comparison and Analysis By accessing historical inspection data from the archive database, a comprehensive anomaly score is given for the same insulator. Perform time series analysis to assess the changing trends of its state.

[0053] Select the most recent N inspection data, use time as the independent variable and P-value as the dependent variable to perform linear regression, and obtain the slope k, which is the trend coefficient.

[0054]

[0055] For the first The timing of the next inspection. The average over time. This is the anomaly score corresponding to the inspection time point. This represents the mean of the anomaly scores. k > 0: indicates that the overall anomaly score P shows an upward trend over time, meaning the insulator condition is deteriorating. k < 0: indicates that the overall anomaly score P shows a downward trend over time, meaning the insulator condition is improving, indicating that the insulator has been maintained or cleaned.

[0056] It can be an insulator comprehensive anomaly score. It can also be a surface damage rating. Electrolytic erosion score Corrosion rating of metal accessories and overall structural anomaly score Specialized scoring, etc.

[0057] when Comprehensive anomaly score for insulators At that time, k reflects the macroscopic trend of the overall health status of the insulator and is the core basis for condition assessment and early warning decision-making.

[0058] when In the scoring of each of the above-mentioned specific items, k reflects the contribution of the corresponding specific item to the change in the overall score. By comparing the absolute values ​​of the trend coefficients of each specific item, the dominant factor X leading to the deterioration of the insulator condition can be identified.

[0059]

[0060] The aforementioned key factor X can directly guide the focus of on-site inspections and clearly indicate the types of anomalies that need to be prioritized in maintenance recommendations.

[0061] S6-3. Predicted Maintenance Date Based on the current comprehensive anomaly score And the corresponding trend coefficient k, to generate a comprehensive anomaly score. The curve of change is used to predict the timeframe for maintenance.

[0062] According to Table 1, maintenance is required for the insulators when P > 0.6. The prediction method is as follows: , In the formula, To predict the time, Comprehensive anomaly score for the current time point The changing trend Comprehensive anomaly score for the current time point The value, The rate of change of the value of k is calculated from the data of the most recent N inspections: , In the formula, For the first The timing of the next inspection. The average over time. For the first The k value during the next inspection. The mean of the k values.

[0063] The following specific data example illustrates this solution: Data from the last 5 inspections of a certain insulator (inspection intervals are 2 months) is as follows: Table 2 Inspection data of a certain insulator

[0064] The current (fifth inspection) trend coefficient has been calculated. , This indicates that the insulators are aging at an accelerated rate. From the above calculation, we can conclude that: , Starting from the current inspection (5th inspection), it is estimated that after approximately 2.75 more inspections, the overall anomaly score of this insulator will be [not specified]. If it exceeds 0.6, it will enter a "critical" state, meaning the predicted maintenance window is 5.5 months.

[0065] This invention also discloses an insulator aging state identification system based on contrastive learning, including an image acquisition module, an identity recognition and data transmission module, an image preprocessing module, a defect identification and analysis module, a data fusion and scoring calculation module, and a state assessment and early warning output module, wherein: The image acquisition module is used to acquire high-definition images of insulators; The identification and data transmission module includes an encoding recognition module and a data transmission module. The encoding recognition module integrates the ZBar QR code recognition algorithm or the PaddleOCR optical character recognition model to read the corresponding encoding information of the insulator; the data transmission module is used to transmit the collected raw images, data and encoding information to the database, and supports the creation of new ID files and the association and matching of old IDs.

[0066] The image preprocessing module includes an instance segmentation module and a perspective transformation module. The instance segmentation module deploys a Mask R-CNN model for insulator image segmentation, umbrella disk mask generation, and contour extraction; the perspective transformation module, based on OpenCV, fits the umbrella disk ellipse and corrects the frontal viewpoint to ensure geometric consistency.

[0067] The defect identification and analysis module includes a semantic segmentation module, an object detection module, and a feature analysis module. The semantic segmentation module uses a U-Net network for binarization segmentation of surface damage areas and segmentation of electro-erosion / carbonization trace areas. The object detection module deploys a YOLOv8 model to locate the bounding boxes of metal attachments and calculate overall structural parameters. The feature analysis module performs color analysis, texture analysis, and size and contour analysis on the image.

[0068] The data fusion and scoring calculation module includes: a scoring calculation unit, a weighted fusion unit, and a historical data management module.

[0069] The scoring calculation unit calculates surface damage scores, electrical erosion scores, metal accessory corrosion scores, and overall structural anomaly scores according to preset formulas. The weighted fusion unit fuses the specific scores based on weight coefficients (α, β, γ, δ) to generate a comprehensive anomaly score, supporting dynamic weight adjustment and machine learning optimization. The historical data management module connects to the database to realize score data storage, historical record retrieval, and time series analysis.

[0070] The status assessment and early warning output module includes a level assessment module, a trend analysis module, and a report generation module. The level assessment module determines the insulator's health status level and corresponding early warning level based on a comprehensive anomaly score and a preset threshold range. The trend analysis module calculates trend coefficients through linear regression, identifies the trend of status changes and the dominant deterioration factors, and predicts the maintenance timeframe.

[0071] The report generation module outputs a comprehensive diagnostic report containing health scores, trend analysis, and maintenance recommendations, supporting inspection analysis and predictive maintenance decisions.

[0072] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying the aging state of insulators based on contrastive learning, characterized in that, Includes the following steps: S1. Acquire high-definition images of insulators; S2. Identify insulators from images and assign them unique identifiers; S3. Preprocess the image, extract the umbrella disk outline and perform perspective transformation to obtain the frontal view image; S4. Perform defect identification on the preprocessed images, including surface damage identification, electrical erosion identification, metal accessory corrosion identification, and overall structural anomaly identification, and generate corresponding specialized scores for each. S5. Weight and merge the scores of each specific item to obtain the comprehensive anomaly score of the insulator; S6. Based on comprehensive anomaly scores and historical data, assess the health status of insulators, analyze trends, and generate early warnings and maintenance recommendations.

2. The insulator aging state identification method based on contrastive learning according to claim 1, characterized in that, In step S2, the insulator and its attached weather-resistant QR code or digital code plate are photographed by a drone, and the encoded information is identified using ZBar or PaddleOCR and matched with the central database or a new file is created.

3. The insulator aging state identification method based on contrastive learning according to claim 1, characterized in that, In step S3, Mask R-CNN is used for instance segmentation to extract the umbrella disk outline, and image deformation is corrected by ellipse fitting and perspective transformation to obtain the frontal viewpoint.

4. The insulator aging state identification method based on contrastive learning according to claim 1, characterized in that, In step S4: the surface damage identification uses U-Net to segment the damaged area and calculates the surface damage score based on the damaged area and location; the electro-erosion identification calculates the score for the electro-erosion powdering area or carbonization area according to the insulator material; the metal accessory corrosion identification uses YOLOv8 to locate the metal component and calculates the corrosion score based on spots, gloss, and integrity; the overall structural anomaly identification uses YOLOv8 to perform axis fitting, curvature calculation, and foreign object detection to generate a structural anomaly score.

5. The insulator aging state identification method based on contrastive learning according to claim 1, characterized in that, The specific scoring methods mentioned in step S5 include: surface damage scoring. Electrolytic erosion score Corrosion rating of metal accessories and overall structural anomaly score Weighted fusion is performed to obtain the comprehensive anomaly score of the insulator. One of the methods is as follows: The calculation method for the comprehensive anomaly score is: ， In the formula, These are the weighting coefficients for surface damage, electrolytic corrosion, metal accessory corrosion, and overall structural anomalies, respectively. .

6. The insulator aging state identification method based on contrastive learning according to claim 1, characterized in that, Step S6 includes: classifying health levels based on the comprehensive anomaly score and generating early warnings and maintenance recommendations; calling historical data for time series analysis, calculating trend coefficients, and identifying dominant deterioration factors; and predicting maintenance time windows based on the current score and trend.

7. The insulator aging state identification method based on contrastive learning according to claim 6, characterized in that, The method for calculating the trend coefficient in step S6 is to select the most recent N inspection data, perform linear regression with time as the independent variable and p-value as the dependent variable, and obtain the slope k, which is the trend coefficient; the calculation method of the trend coefficient is as follows: ， For the first The timing of the next inspection. The average over time. This is the anomaly score corresponding to the inspection time point. This represents the mean of the abnormality scores.

8. The insulator aging state identification method based on contrastive learning according to claim 6, characterized in that, The method for predicting the maintenance time window in step S6 is to generate a change curve for the comprehensive anomaly score P based on the current comprehensive anomaly score P and the corresponding trend coefficient k, and then predict the time period required for maintenance. The calculation method is as follows: ， In the formula, To predict the time, This represents the trend of the comprehensive anomaly score P at the current point in time. The value of the comprehensive anomaly score P at the current time point. The rate of change of the value of k is calculated from the data of the most recent N inspections: ， In the formula, For the first The timing of the next inspection. The average over time. For the first The k value during the next inspection. The mean of the k values.

9. A system for identifying the aging state of insulators based on contrastive learning as described in claim 1, characterized in that, include: The image acquisition module is used to acquire high-definition images of insulators; The identification and data transmission module is used to identify the insulator and transmit data. The image preprocessing module is used for image segmentation and viewpoint correction. The defect identification and analysis module is used to identify various defects and generate specific scores; The data fusion and scoring calculation module is used to calculate the comprehensive anomaly score and manage historical data; The status assessment and early warning output module is used for health assessment, trend analysis, and report generation.

10. The insulator aging state identification system based on contrastive learning according to claim 9, characterized in that, The identity recognition and data transmission module includes an encoding recognition module and a data transmission module. The encoding recognition module integrates the ZBar QR code recognition algorithm or the PaddleOCR optical character recognition model to read the corresponding encoding information of the insulator. The data transmission module is used to transmit the acquired raw images, data, and encoded information to the database, and supports the creation of new ID files and their association and matching with old IDs; The image preprocessing module includes an instance segmentation module and a perspective transformation module. The instance segmentation module deploys a Mask R-CNN model for insulator image segmentation, umbrella disk mask generation, and contour extraction. The perspective transformation module, based on OpenCV, fits the umbrella disk ellipse and corrects the frontal view to ensure geometric consistency. The defect identification and analysis module includes a semantic segmentation module, an object detection module, and a feature analysis module. The semantic segmentation module is equipped with a U-Net network and is used for binarization segmentation of surface damage areas and segmentation of electro-erosion / carbonization trace areas, respectively. The object detection module deploys a YOLOv8 model to locate the bounding box of metal attachments and calculate the overall structural parameters. The feature analysis module is used for color analysis, texture analysis, and size and contour analysis of images. The data fusion and scoring calculation module includes: a scoring calculation unit, a weighted fusion unit, and a historical data management module; The scoring calculation unit calculates the surface damage score, electrical erosion score, metal accessory corrosion score, and overall structural anomaly score according to the preset formula. The weighted fusion unit fuses the specific scores based on the weight coefficients (α, β, γ, δ) to generate a comprehensive anomaly score. It supports dynamic adjustment of weights and machine learning optimization. The historical data management module connects to the database to realize the storage of scoring data, retrieval of historical records, and time series analysis. The status assessment and early warning output module includes a level assessment module, a trend analysis module, and a report generation module. The level assessment module determines the insulator's health status level and corresponding early warning level based on the comprehensive anomaly score and preset threshold range. The trend analysis module calculates trend coefficients through linear regression, identifies the trend of status changes and the dominant deterioration factors, and predicts the maintenance timeframe. The report generation module outputs a comprehensive diagnostic report containing health scores, trend analysis, and maintenance recommendations, supporting inspection analysis and predictive maintenance decisions.

Images