Fault diagnosis method and system for point cloud data of electric power tower

Abstract

The invention relates to the related technical field of data analysis and processing, in particular to a power tower point cloud data fault diagnosis method and system, and the method comprises the steps: collecting power tower point cloud related core component fixed local geometric features and related architecture component fixed global features, and combining with an unmanned plane image fusion point cloud to obtain a power tower point cloud fault diagnosis result; and comparing with the standard characteristics to verify inclination, part damage and insulator deterioration faults, and carrying out failure reminding. The technical problems that for hidden faults such as insulator degradation and part corrosion, a single data source is prone to noise interference, and fault failure reminding is difficult to effectively carry out are solved, unmanned aerial vehicle visual images and point cloud data are combined, the defects of the single data source in detail representation and space structure are overcome, and the fault reminding efficiency is improved. According to the method, standard geometric features associated with the model and parameters of the tower are established, inclination fault verification, component damage fault verification and insulator deterioration fault verification are carried out on the electric power tower, and the reliability of fault failure reminding is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of data analysis and processing technology, specifically to a fault diagnosis method and system for point cloud data of power poles. Background Technology

[0002] As a key supporting structure for power transmission networks, the safe and stable operation of power poles is crucial to ensuring power supply. However, long-term exposure to complex natural environments and mechanical stresses can easily lead to faults such as tilting, component damage, and insulator deterioration. These not only threaten the safety of the power grid but may also cause serious accidents such as large-scale power outages and fires.

[0003] Current power pole fault diagnosis technology has significant limitations. Detection methods based on single sensors (such as cameras and inclinometers) are insufficient in identifying hidden faults, cannot fully reflect the overall structural health of the pole, and are difficult to accurately identify early faults such as minor deformation and corrosion. Furthermore, the real-time performance and intelligence level of fault diagnosis are insufficient, failing to meet the urgent needs of power grid operation and maintenance for fault early warning and life prediction.

[0004] In summary, existing technologies have technical problems such as difficulty in effectively alerting to hidden faults like insulator deterioration and component corrosion, susceptibility of single data sources to noise interference, and difficulty in providing fault failure warnings. Summary of the Invention

[0005] This application provides a fault diagnosis method and system for power pole point cloud data, aiming to solve the technical problems in the prior art where hidden faults such as insulator deterioration and component corrosion are difficult to effectively detect due to noise interference from a single data source.

[0006] In view of the above problems, the technical solution to achieve the present application is as follows: This application provides a fault diagnosis method for power pole point cloud data. The method includes: collecting point cloud data from power poles; associating this data with core components of the power pole to determine local geometric features, where the core components include angle steel and steel pipes corresponding to the tower body, and crossarms and grounding frames corresponding to the tower head; associating the power pole structure components, including the tower body, tower head, and connecting members, to determine global geometric features; combining the local and global geometric features based on image data acquired by a UAV, and using voxel grid downsampling to obtain fused point cloud data; comparing and analyzing the fused point cloud data with standard geometric features in a data storage unit to confirm feature similarity and deviation values; performing tilt fault verification, component damage fault verification, and insulator deterioration fault verification on the power pole; configuring a high-dimensional fault characteristic tensor; and providing failure alerts based on the physical characteristics of the connection method limited by the connecting members.

[0007] Preferably, based on the high-dimensional fault characteristic tensor, the fault failure components and fault evolution index are determined; according to the fault failure components and fault evolution index, the feasible region of the fault verification database is updated, the feasible region is mapped to the set of fault feature vectors obtained by the search, and the fault verification database is mapped to the tilt fault verification interval, the component damage fault verification interval, and the insulator deterioration fault verification interval; at the same time, the weighted entropy value of the fault evolution index and the diagnostic accuracy is used as the reward function to construct a reinforcement learning framework to dynamically optimize the diagnostic strategy. When the confidence of the detected feature vector set is lower than a preset threshold, the contrastive learning unit is triggered to enhance the fault feature discrimination.

[0008] Preferably, the theoretical center axis of the power pole body and the theoretical center axis of the tower head are determined; the tilt of the power pole is evaluated by the angle and offset distance between the actual center axis and the theoretical center axis of the tower body and the theoretical center axis of the tower head; the tilt of the power pole is compared with a preset tilt threshold to confirm the tilt fault verification result.

[0009] Preferably, based on the core components of the power pole, the pole point cloud data is associated and segmented to obtain multiple segmented point cloud data sets. This includes setting a density threshold and a neighborhood radius, traversing the pole point cloud data, collecting the number of points in the neighborhood of each point, and if the point density is greater than the density threshold, it is recorded as a core point; classifying the core point and the remaining points within the neighborhood radius into the same category to obtain multiple segmented point cloud data sets associated with the core components of the power pole; formulating a covariance matrix based on the multiple segmented point cloud data sets; obtaining the eigenvalues ​​and eigenvectors of the covariance matrix, sorting them according to the eigenvalues ​​from largest to smallest, selecting the eigenvectors corresponding to the first N eigenvalues, and setting an eigenvector matrix; and determining the local geometric features based on the eigenvector matrix.

[0010] Preferably, point cloud data of similar power poles under different operating conditions are introduced, and similar fused point cloud data corresponding to the core components and structural components of similar power poles are extracted; based on the similar fused point cloud data, the corrosion coordinates of angle steel and the deformation coordinates of steel pipe are identified, and the corrosion depth and deformation parameters are extracted; the corrosion depth and deformation parameters are input into a pre-trained fatigue life prediction model to dynamically evaluate the remaining service life of the core components of the power pole.

[0011] Preferably, a standard feature vector library is established based on the standard geometric features; the standard feature vector library is stored in the data storage unit; and the standard feature vectors are associated and combined with the corresponding tower models and parameter information according to the similar fused point cloud data to obtain standard geometric features.

[0012] Preferably, based on the standard feature vector library, multiple feature vectors to be compared are identified, and the cosine similarity and Euclidean distance deviation value between the first feature vector to be compared and the associated standard feature vector are obtained, wherein the first feature vector to be compared is any one of the multiple feature vectors to be compared; if the cosine similarity is lower than a preset similarity threshold or the Euclidean distance deviation value exceeds the dynamic compensation parameter, an anomaly alert is triggered, and the fault component type is located based on the deviation direction.

[0013] Preferably, based on point cloud samples of core components of power poles with different degrees of damage, the damage type and degree level are labeled to construct a multi-dimensional damage feature vector; based on the multi-dimensional damage feature vector, a support vector machine is used to train a classification model with cross-validation to form a diagnostic model that can identify damage to typical components.

[0014] Preferably, a subset of the point cloud of the insulator string is extracted to obtain the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt of the insulator string; based on the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt of the insulator string, a spectral analysis operation is performed in conjunction with the image data acquired by the UAV vision, and the insulator degradation level is determined by fuzzy evaluation method; wherein, the spectral analysis operation includes an ultraviolet corona detection operation node, an infrared thermal imaging analysis operation node, and a visible light band spectral analysis operation node.

[0015] In another aspect, this application provides a fault diagnosis system for power pole point cloud data, wherein the system includes: a data acquisition module, used to acquire power pole point cloud data, and associate it with the core components of the power pole to determine local geometric features, wherein the core components of the power pole include angle steel and steel pipes corresponding to the tower body, and crossarms and grounding frames corresponding to the tower head; a feature extraction module, used to determine global geometric features by associating the power pole structure components including the tower body, tower head and connecting rods; a feature combination module, used to combine the local geometric features and global geometric features according to image data acquired by UAV vision, and use voxel grid downsampling to obtain fused point cloud data; and a fault verification module, used to compare and analyze the fused point cloud data with standard geometric features in the data storage unit, confirm feature similarity and deviation values, perform tilt fault verification, component damage fault verification, and insulator deterioration fault verification on the power pole, configure a high-dimensional fault characteristic tensor, and provide failure alerts in combination with the physical characteristics of the connection method limited by the connecting rods.

[0016] In summary, one or more technical solutions provided in this application achieve the separation of local geometric features of core components from global geometric features of architectural components, and combine UAV visual images and point cloud data to make up for the deficiencies of a single data source in terms of detailed representation and spatial structure. This establishes standard geometric features associated with tower models and parameters, and performs tilt fault verification, component damage fault verification, and insulator deterioration fault verification on power towers, thus ensuring the reliability of fault failure alerts. Attached Figure Description

[0017] Figure 1 This application provides a flowchart illustrating a method for fault diagnosis using point cloud data of power poles.

[0018] Figure 2 This application provides a structural schematic diagram of a fault diagnosis system for power pole point cloud data.

[0019] Explanation of reference numerals in the attached diagram: Data acquisition module M100, feature extraction module M200, feature combination module M300, fault verification module M400. Detailed Implementation

[0020] Example 1: The present application will be described in detail below with reference to the accompanying drawings, as follows... Figure 1 As shown, this application provides a fault diagnosis method for power pole point cloud data, wherein the method includes: S1: Collect point cloud data of the power poles and determine local geometric features by associating them with the core components of the power poles. The core components of the power poles include the angle steel and steel pipes corresponding to the tower body, and the crossarms and grounding wires corresponding to the tower head. S2: Determine the global geometric features by associating the power pole structure components, including the tower body, tower head, and connecting members. S3: Based on the image data collected by the UAV, combine the local geometric features and global geometric features, and obtain fused point cloud data by using voxel grid downsampling.

[0021] Specifically, pole point cloud data refers to a large amount of three-dimensional point data obtained by collecting data on power poles using laser scanning technology. This data can accurately reflect the surface shape and structural information of the pole. Among them, the core components of power poles refer to the parts that play a key supporting and connecting role in the pole, including the angle steel and steel pipes of the tower body, the crossarms of the tower head and the grounding frame, etc. The integrity and stability of these components are directly related to the structural safety of the entire pole.

[0022] Local geometric features refer to the geometric shape characteristics of these core components in local areas, such as the cross-sectional shape of angle steel and the degree of bending of steel pipes, reflecting the local structural details of the core components; while power pole structure components cover the various parts that constitute the entire pole structure, such as the tower body, tower head and connecting rods. Global geometric features refer to the geometric features of the entire pole structure at the overall level, such as the overall height of the pole, the tilt angle of the tower body, and the relative positional relationship between the tower head and the tower body, reflecting the overall form and spatial structure of the pole.

[0023] Combining refers to integrating local and global geometric features, that is, combining the local features of core components with the global features of architectural components to form a more comprehensive geometric description of the tower. Voxel mesh downsampling refers to dividing point cloud data into individual voxels (which can be understood as small cubes in three-dimensional space), and then downsampling within each voxel, that is, selecting a certain number of points to represent the data within that voxel, thereby reducing the total amount of point cloud data while retaining the main geometric features, resulting in fused point cloud data. That is, after downsampling processing, a point cloud dataset that integrates local and global geometric features is obtained. The fused point cloud data contains both the detailed features of core components and reflects the macroscopic structural features of the entire tower, providing a concise and effective data foundation for subsequent fault diagnosis.

[0024] In one feasible implementation, point cloud data is collected from power poles. A laser scanning device is used to perform a full-range scan of the poles, which can acquire high-density point cloud data, with the number of data points reaching millions or even tens of millions. Based on the structural characteristics of the core components of the power poles, local geometric features related to the core components are extracted from the collected pole point cloud data. For example, by using specific algorithms to identify features such as the edge contour of angle steel and the curvature change of steel pipes, the edge points of angle steel in a certain pole can be accurately extracted. The shape formed by these edge points is the local geometric feature.

[0025] Simultaneously, based on the point cloud data of the entire power pole, the global geometric features of the power pole structure components are analyzed, such as the total height of the pole being 30 meters and the tilt angle of the tower being 2 degrees. These local and global geometric features are combined, that is, the local feature information of the core components is integrated into the global feature framework of the entire pole. The combined point cloud data is processed using voxel mesh downsampling technology. After downsampling, the amount of data is reduced, resulting in fused point cloud data.

[0026] By extracting local and global geometric features from the associated core components and architectural components and performing downsampling fusion, not only is the key geometric information of the tower preserved, but the data scale is also effectively reduced, reducing the amount of computation and resource consumption in the subsequent fault diagnosis process, and improving the operating efficiency and diagnosis speed of the entire fault diagnosis system.

[0027] S4: Using the fused point cloud data, compare and analyze it with the standard geometric features in the data storage unit to confirm the feature similarity and deviation value. Perform tilt fault verification, component damage fault verification, and insulator deterioration fault verification on the power pole. Configure a high-dimensional fault characteristic tensor and combine it with the physical characteristics of the connection method under the limitation of the connecting rod to provide failure reminder.

[0028] Specifically, comparative analysis refers to comparing the fused point cloud data with standard geometric features pre-stored in the data storage unit one by one to find the similarities and differences between the two; feature similarity is calculated by a specific algorithm to determine the degree of matching between the fused point cloud data and the standard geometric features, usually expressed as a percentage or score, reflecting the closeness of the two in terms of geometric features; deviation value measures the degree of difference between the fused point cloud data and the standard geometric features, which can be expressed numerically, with a smaller deviation value indicating that the two are closer.

[0029] The data storage unit refers to a database or memory specifically designed to store standard geometric features, capable of saving the standard geometric features of power poles of different models and parameters for comparative analysis; the high-dimensional fault characteristic tensor is a high-dimensional data structure used to comprehensively represent various fault characteristics of power poles, capable of integrating feature information of different fault types and components, providing a comprehensive quantitative basis for fault diagnosis; the physical characteristics of the connection method under the constraint of connecting rods refer to the mechanical properties and physical constraints determined by the connection form (such as welding, bolt connection, etc.) of the connecting rods in the power pole, which are closely related to the fault mode and can be used for failure analysis and early warning.

[0030] In one feasible implementation, the fused point cloud data is compared and analyzed with the standard geometric features in the data storage unit. The difference between the actual state and the standard state of the power pole is evaluated by calculating the feature similarity and deviation value. For example, in the tilt fault verification, if the feature similarity is lower than the set similarity threshold (such as 90%), or the deviation value exceeds the set deviation range (such as the deviation angle exceeding 3 degrees or the deviation distance exceeding 5 centimeters), then it is determined that there is a tilt fault.

[0031] Similarly, component damage faults and insulator deterioration faults are verified, and the existence of faults is determined based on the corresponding similarity and deviation values. Then, the tilt fault characteristics, component damage fault characteristics, and insulator deterioration fault characteristics are integrated into a high-dimensional fault characteristic tensor. The dimensions of the high-dimensional fault characteristic tensor may include fault type (tilt, component damage, insulator deterioration), tower location (tower body, tower head, connecting rods), fault severity level (minor, moderate, severe), etc. The value range of each dimension is determined according to the actual detection standard. For example, the fault severity level can be divided into 1-5 levels.

[0032] Failure alerts are triggered by combining the physical characteristics of the connecting members (such as connection strength and fatigue life). For example, a failure alert is triggered when the connecting members are bolted and the calculated remaining fatigue life is less than 1000 stress cycles. Through precise feature comparison and multi-dimensional fault integration, the above steps enable comprehensive and efficient diagnosis of various fault types in power poles, significantly improving the accuracy and reliability of fault diagnosis. This provides solid data support for the operation and maintenance decisions of power poles, effectively ensuring the safe and stable operation of the power system.

[0033] Furthermore, the method of this application includes: Based on the high-dimensional fault characteristic tensor, the fault failure components and fault evolution index are determined. According to the fault failure components and fault evolution index, the feasible region of the fault verification database is updated. The feasible region is mapped to the searched set of fault feature vectors, and the fault verification database is mapped to the tilt fault verification interval, component damage fault verification interval, and insulator deterioration fault verification interval. Simultaneously, using the weighted entropy value of the fault evolution index and diagnostic accuracy as the reward function, a reinforcement learning framework is constructed to dynamically optimize the diagnostic strategy. When the confidence level of the detected feature vector set is lower than a preset threshold, a contrastive learning unit is triggered to enhance the fault feature discrimination.

[0034] Specifically, the high-dimensional fault characteristic tensor is used to comprehensively represent various fault characteristics of power poles, including but not limited to tilting, component damage, and insulator deterioration; the fault failure component refers to the specific pole component identified as having a fault risk during the diagnosis process, such as angle steel, steel pipe, crossarm, and insulator; the fault evolution index is a quantitative indicator used to measure the development of a fault over time or under changing operating conditions, reflecting the severity and trend of the fault; the feasible region refers to the reasonable range set of fault feature vectors defined in the fault verification database, mapped to a set of fault feature vectors obtained from historical data or search algorithms, used to limit the acceptable range of feature parameters during the fault diagnosis process.

[0035] The fault verification database is used to store various fault verification standards, parameter ranges, and related feature information, including tilt fault verification ranges, component damage fault verification ranges, and insulator deterioration fault verification ranges, providing a benchmark reference for fault diagnosis; mapping refers to establishing a correspondence between the feasible domain and the set of fault feature vectors, and between the fault verification database and the verification range; the weighted entropy value is a quantitative indicator that comprehensively considers the importance of different fault features and is used to evaluate the merits of the diagnostic strategy.

[0036] Reinforcement learning frameworks are machine learning methods based on trial-and-error learning and reward signals, used to train agents to make optimal decisions in complex environments to maximize cumulative rewards; contrastive learning units are technical modules that enhance feature representation capabilities by comparing the similarity and differences between samples, aiming to improve the model's ability to distinguish different fault features; confidence refers to the degree of certainty that a set of feature vectors belongs to a certain fault type. When it is lower than a preset threshold, it indicates that the uncertainty of the current diagnostic result is high.

[0037] In one feasible implementation, based on the high-dimensional fault characteristic tensor, the fault failure component and fault evolution index are determined by analyzing the fault characteristic information therein. For example, if a certain dimension of the high-dimensional fault characteristic tensor corresponds to the deterioration characteristics of the insulator, and its fault evolution index reaches 0.8 (indicating a high degree of severity and affecting power transmission efficiency), then the insulator can be determined to be a fault failure component.

[0038] Based on the failed components and the failure evolution index, the feasible domain of the fault verification database is updated. If the original feasible domain includes the fault feature vector set A, B, and C, the updated domain may narrow or expand the range. For example, the fault verification range for component damage may be adjusted from 0-100% of the damage level to 0-60% (if the damage level exceeds 60%, repair or replacement is required), thereby more accurately defining the fault range.

[0039] Using the weighted entropy value of fault evolution index and diagnostic accuracy as the reward function, a reinforcement learning framework is constructed to dynamically optimize the diagnostic strategy. For example, when the diagnostic accuracy reaches 90%, the weight of the fault evolution index is set to 0.7 (emphasizing the impact of fault severity), and the weight of the diagnostic accuracy is set to 0.3. The resulting weighted entropy value can be used as a reward signal to guide the reinforcement learning framework to adjust the diagnostic strategy parameters, so that the diagnostic model learns the optimal diagnostic method through continuous trial and error.

[0040] When the confidence level of the detected feature vector set falls below a preset threshold (e.g., confidence level below 85%), a contrastive learning unit is triggered to enhance the distinguishability of fault features. Specifically, through contrastive learning, the distinguishability of similar fault features (e.g., angle steel with different degrees of damage) is improved, enabling the model to more accurately distinguish different fault types, thereby enhancing the reliability of diagnosis. In the above steps, by dynamically adjusting the diagnostic strategy and enhancing feature distinguishability, the accuracy and adaptability of fault diagnosis can be effectively improved, increasing the diagnostic accuracy rate and ensuring the stable operation of the power system.

[0041] Furthermore, the method for verifying tilting faults of power poles in this application includes: Determine the theoretical center axis of the power pole body and the theoretical center axis of the tower head; assess the degree of tilt of the power pole by measuring the angle and offset distance between the actual center axis and the theoretical center axis of the tower body and the theoretical center axis of the tower head; compare the degree of tilt of the power pole with a preset tilt threshold to confirm the tilt fault verification result.

[0042] Specifically, the theoretical center axis refers to the ideal center axis of the tower body and tower head determined according to the design specifications or standard geometric characteristics of power poles. It describes the central symmetry axis that the tower body and tower head should maintain in an ideal state from a geometric perspective and can be used as a benchmark reference for assessing the tower tilt. The actual center axis refers to the actual geometric center axis of the tower body and tower head calculated based on the actual collected tower point cloud data, reflecting the real situation of the tower under actual operating conditions. The angle between the theoretical center axis of the tower body and the theoretical center axis of the tower head refers to the angular deviation between the actual center axis and the theoretical center axis, which can be used to measure the degree of tower tilt. The offset distance refers to the spatial distance deviation between the actual center axis and the theoretical center axis, which can also be used to assess the tower tilt status. The tilt fault verification result refers to the analysis of the angle and offset distance between the actual center axis and the theoretical center axis, and the determination of whether the tower has a tilt fault and the severity of the tilt based on a preset tilt threshold.

[0043] In one feasible implementation, the theoretical center axis of the power pole body and the theoretical center axis of the tower head are determined to reflect the difference between the actual geometric characteristics of the pole and the standard geometric characteristics. Further, the theoretical center axis of the pole body is determined by combining the geometric center and direction vector of the pole body point cloud data with the theoretical geometric shape of the pole body in the standard geometric characteristics; similarly, the theoretical center axis of the tower head is determined. Next, the tilt degree of the power pole is assessed by the angle and offset distance between the actual center axis (the actual center axis is the geometric center axis calculated based on the pole point cloud data) and the theoretical center axis. Through the analysis of the pole center axis, the tilt degree of the pole is accurately quantified. Based on geometric characteristics and the analysis method comparing actual and theoretical values, tilt faults of the pole can be effectively identified, improving the accuracy and reliability of fault diagnosis, avoiding potential tower collapse accidents caused by tilt, and ensuring the safe and stable operation of power poles.

[0044] Furthermore, the method involves collecting point cloud data of power poles and associating it with the core components of the power poles to determine local geometric features. The method also includes: Based on the core components of the power pole, the pole point cloud data is associated and segmented to obtain multiple segmented point cloud data sets. This includes setting a density threshold and a neighborhood radius, traversing the pole point cloud data, collecting the number of points in the neighborhood of each point, and if the point density is greater than the density threshold, it is recorded as a core point; classifying the core point and the remaining points within its neighborhood radius into the same category to obtain multiple segmented point cloud data sets associated with the core components of the power pole; formulating a covariance matrix based on the multiple segmented point cloud data sets; obtaining the eigenvalues ​​and eigenvectors of the covariance matrix, sorting them according to the eigenvalues ​​from largest to smallest, selecting the eigenvectors corresponding to the first N eigenvalues, and setting an eigenvector matrix; and determining the local geometric features based on the eigenvector matrix.

[0045] Specifically, correlation segmentation refers to segmenting the parts belonging to different core components from the collected point cloud data of power poles based on the structural and location information of the core components of the poles, resulting in multiple segmented point cloud data sets. Each set corresponds to a specific core component, such as the angle steel of the tower body or the crossarm of the tower head. The covariance matrix can reflect the correlation and dispersion between data points. By calculating the covariance matrix of the segmented point cloud data sets, the distribution characteristics of the point cloud data in different dimensions can be captured.

[0046] Eigenvalues ​​and eigenvectors are concepts in linear algebra. For a covariance matrix, eigenvalues ​​represent the variance of the data in the direction of the corresponding eigenvector, i.e., the degree of dispersion of the data in that direction. When processing point cloud data, sorting the eigenvalues ​​from largest to smallest and selecting the eigenvectors corresponding to the top N eigenvalues ​​can determine the main direction of change of the data. The matrix formed by these eigenvectors is called the eigenvector matrix, which represents the main geometric feature direction of the point cloud data. Local geometric features refer to the features that can reflect the local shape and structural features of the core components, as determined by the eigenvector matrix, such as the length direction of angle steel and the width direction of crossarms.

[0047] In one feasible implementation, based on the structural characteristics of the core components of the power pole tower, such as the shape and position of the grounding wire frame (angle steel, steel pipe, crossarm, etc.), the point cloud data belonging to these core components is segmented from the pole tower point cloud data; a covariance matrix is ​​calculated for each segmented point cloud data set. The dimension of the covariance matrix is ​​usually related to the spatial dimension of the point cloud data. For three-dimensional point cloud data, the covariance matrix is ​​a 3×3 matrix; the eigenvalues ​​and eigenvectors of the covariance matrix are obtained and sorted in descending order of eigenvalues.

[0048] For example, the three eigenvalues ​​obtained are λ1=120, λ2=50, and λ3=30, and the corresponding eigenvectors are v1, v2, and v3, respectively. Select the eigenvectors corresponding to the first N (assuming N=2) eigenvalues, namely v1 and v2, and set the eigenvector matrix as [v1, v2]. Determine the local geometric features based on the eigenvector matrix [v1, v2]. Here, eigenvector v1 represents the main extension direction of the angle steel, and v2 represents the secondary change direction of the angle steel. These two directions together determine the local geometric features of the angle steel.

[0049] By performing correlation segmentation and feature analysis on the point cloud data of the power pole, the local geometric features of the core components can be accurately extracted. These local geometric features of the core components of the power pole will be combined with global geometric features for subsequent fusion point cloud data generation and fault verification, improving the accuracy and reliability of fault diagnosis. Furthermore, in subsequent component damage fault verification, by analyzing these local geometric features, it is possible to more accurately detect whether angle steel has deformed, broken, or other faults, thereby timely identifying potential safety hazards and ensuring the safe operation of the power pole.

[0050] Based on the core components of the power pole, the pole point cloud data is segmented and correlated to obtain multiple segmented point cloud data sets. Specifically, the density threshold refers to a pre-set point density standard used to distinguish between core and non-core areas in the point cloud data, typically expressed as points per cubic centimeter or per cubic meter. The neighborhood radius refers to the radius of the surrounding area centered on a given point in the point cloud data, used to collect other points within that neighborhood, expressed as centimeters or meters. Traversal refers to sequentially visiting each point in the pole point cloud data and analyzing and processing each point.

[0051] Point number density refers to the number of points per unit volume within the neighborhood of a given point, reflecting the point density of that area. Core points are those with a point number density greater than a set density threshold. These points are typically located in dense areas of point cloud data and are the locations of the core components of the power pole. Classifying points into the same category means grouping the core points and the remaining points within their neighborhood radius into the same category, that is, considering these points to belong to the same core component or the same structural region. This results in multiple segmented point cloud data sets associated with the core components of the power pole. Each segmented point cloud data set contains the point cloud data of a core component and its adjacent area.

[0052] In one feasible implementation, a suitable density threshold and neighborhood radius are set; then, each point in the tower point cloud data is traversed, and for each point, the number of points within its neighborhood radius is collected, and the point density is calculated; if the point density in the neighborhood of a certain point is greater than the density threshold, it is marked as a core point; each core point and the remaining points within its neighborhood radius are divided into the same category to obtain multiple segmented point cloud data sets. For example, five segmented point cloud data sets are successfully segmented from the tower point cloud data, and each set corresponds to a core component of a power tower, such as the angle steel of the tower body and the crossarm of the tower head.

[0053] By setting density thresholds and neighborhood radii, it is possible to effectively extract the point cloud data set of core components from complex point cloud data. This adapts to core components of different shapes and sizes. The accurately segmented angle steel point cloud data set can be used for subsequent local geometric feature extraction, thereby detecting whether the angle steel has rusted, deformed, or other faults, thus improving the accuracy and efficiency of point cloud data segmentation.

[0054] Furthermore, by combining the local and global geometric features and using voxel mesh downsampling to obtain fused point cloud data, the method of this application also includes: Point cloud data of similar power poles under different working conditions are introduced, and similar fused point cloud data corresponding to the core components and structural components of similar power poles are extracted. Based on the similar fused point cloud data, the corrosion coordinates of angle steel and the deformation coordinates of steel pipe are identified, and the corrosion depth and deformation parameters are extracted. The corrosion depth and deformation parameters are input into a pre-trained fatigue life prediction model to dynamically evaluate the remaining service life of the core components of the power pole.

[0055] Specifically, operating conditions refer to the working status and environmental conditions of a machine, equipment, or system during operation. Point cloud data of the same type of power pole under different operating conditions refers to the collection of point cloud data obtained by scanning and collecting the same type of power pole (similar in structure and purpose) under different operating states and environmental conditions (such as weather, load, etc.).

[0056] Similar fused point cloud data refers to the data obtained by fusing the point cloud data of the core components and structural components of similar power poles under different operating conditions according to certain rules. The purpose is to integrate the feature information under different operating conditions to more comprehensively reflect the geometric state of the power poles. Furthermore, the coordinate positions of specific targets (such as angle steel corrosion and steel pipe deformation) and related corrosion depth and deformation parameters are identified and obtained from the fused point cloud data. These parameters are important indicators for measuring the degree of damage to the core components of the poles.

[0057] A pre-trained fatigue life prediction model refers to a machine learning model that is pre-trained using a large amount of historical data (including data on the geometric features, corrosion, deformation, and corresponding service life records of the tower). It can dynamically assess the remaining service life of the core components of the power tower (i.e., the length of time it can still be used safely under the current damage conditions) based on the input corrosion depth and deformation parameters.

[0058] In one feasible implementation, core components and structural components are extracted from the point cloud data of similar power poles obtained under different working conditions (e.g., sunny days, rainy days, high temperatures, low temperatures, and different loads). Based on these fused point cloud data, image recognition and data processing technologies are used to identify the specific coordinate points of angle steel corrosion and the coordinate areas of steel pipe deformation, and to extract the corrosion depth at the corrosion site (e.g., by comparing the height information in the point cloud data, it is found that the corrosion depth of the angle steel at a certain location reaches 5 mm) and the deformation parameters of the steel pipe.

[0059] These corrosion depth and deformation parameters are input into a pre-trained fatigue life prediction model. Based on a large amount of historical data and machine learning algorithms, the fatigue life prediction model can dynamically assess the remaining service life of the core components of power poles. By introducing multi-condition data, the comprehensiveness and accuracy of the assessment are improved, thereby better ensuring the safe and stable operation of power poles, providing a scientific basis for the maintenance and replacement of power poles, and effectively improving the foresight and efficiency of operation and maintenance.

[0060] Furthermore, by comparing and analyzing the fused point cloud data with the standard geometric features in the data storage unit to confirm the feature similarity and deviation values, the method of this application includes: A standard feature vector library is established based on the standard geometric features; the standard feature vector library is stored in the data storage unit; based on the fused point cloud data of the same type, the standard feature vectors are associated and combined with the corresponding tower type and parameter information to obtain standard geometric features.

[0061] Specifically, standard geometric features refer to the geometric features determined according to the design specifications or standard models of power poles. They refer to the geometric shape and size of power poles under ideal conditions, including the geometric information of core components and structural components under different pole models and parameters. The standard feature vector library refers to the collection stored after standard geometric features are converted into feature vectors. The feature vectors are extracted from standard geometric features through specific algorithms and can effectively characterize the geometric characteristics of the poles.

[0062] Establishing a standard feature vector library refers to the process of using computer algorithms to process standard geometric features, convert them into feature vectors, and store them. This library is stored in a data storage unit, which is a device or system used to store data, such as a database or hard drive. Similar fused point cloud data refers to a data set obtained by fusing point cloud data of the core components and structural components of similar power poles under different operating conditions. It contains geometric feature information of similar poles under different conditions. Association and combination refers to binding and integrating standard feature vectors with corresponding pole models and parameter information to form a complete dataset containing geometric features and their corresponding pole models and parameters, thus obtaining standard geometric features. These features can be used for subsequent comparative analysis and fault diagnosis.

[0063] In one feasible implementation, standard geometric features are obtained based on standard design drawings or models of power poles. These standard geometric features cover various pole models and parameters. Specifically, for different types of poles such as common 110kV straight-line towers and 220kV tension towers, standard geometric features of the tower body, tower head, angle steel, steel pipe, crossarm, and other grounding wire frames are extracted, including information such as shape, size, and relative position. These standard geometric features are then converted into feature vectors to obtain feature vectors that can characterize the geometric features. These feature vectors are then stored in a data storage unit to establish a standard feature vector library.

[0064] Based on similar fused point cloud data, standard feature vectors are associated and combined with corresponding tower models and parameter information to obtain standard geometric features. The corresponding feature vectors are extracted from the standard feature vector library and combined with the tower model and parameter information to generate the standard geometric features of that tower model. By establishing a standard feature vector library and associating and combining it to obtain standard geometric features, the fault diagnosis system can accurately identify the differences between the actual geometric features of the tower and the standard features, thereby achieving accurate verification and diagnosis of tilting faults, component damage faults, insulator deterioration faults, etc., improving the accuracy and reliability of the entire fault diagnosis system.

[0065] Furthermore, the method of this application includes: According to the standard feature vector library, multiple feature vectors to be compared are identified, and the cosine similarity and Euclidean distance deviation value between the first feature vector to be compared and the associated standard feature vector are obtained. The first feature vector to be compared is any one of the multiple feature vectors to be compared. If the cosine similarity is lower than the preset similarity threshold or the Euclidean distance deviation value exceeds the dynamic compensation parameter, an anomaly alert is triggered, and the fault component type is located based on the deviation direction.

[0066] Specifically, cosine similarity measures the similarity between the feature vector to be compared and the standard feature vector in the direction of geometric features. Its value ranges from -1 to 1. The closer the value is to 1, the more similar the directions are, and the closer the value is to -1, the more opposite the directions are. Euclidean distance deviation refers to the distance between the feature vector to be compared and the standard feature vector in Euclidean space. The smaller the distance, the closer the two are in the geometric feature space. Dynamic compensation parameter is a deviation allowable range that is dynamically adjusted according to actual working conditions, data fluctuations, and other factors. It is used to determine whether the Euclidean distance deviation value exceeds the normal range. Deviation direction refers to the direction of difference between the feature vector to be compared and the standard feature vector. By analyzing the deviation direction, it is possible to locate which component type of the tower has a fault. For example, if the deviation direction points to the feature dimension related to angle steel, the faulty component type may be angle steel.

[0067] In one feasible implementation, the feature vector to be compared is identified based on the standard feature vector in the standard feature vector library. Furthermore, the standard feature vector is associated with the geometric features corresponding to different tower models and parameters. The cosine similarity and Euclidean distance deviation values ​​between each feature vector to be compared and the associated standard feature vector are obtained. The cosine similarity is compared with a preset similarity threshold, and the Euclidean distance deviation value is compared with a dynamic compensation parameter.

[0068] If the cosine similarity is below the threshold or the Euclidean distance deviation exceeds the dynamic compensation parameter, an anomaly alert is triggered, and the faulty component type is located based on the deviation direction. Through comparative analysis with standard feature vectors, the differences between the actual geometric features of the tower and the standard features can be accurately identified, enabling timely alerts and precise fault location. This timely notification effectively prevents further escalation of the fault, ensures the safe and stable operation of power towers, and improves the accuracy and reliability of fault diagnosis.

[0069] Furthermore, the method for verifying component faults on power poles in this application includes: Based on point cloud samples of core components of power poles with different degrees of damage, the damage type and severity level are labeled, and a multi-dimensional damage feature vector is constructed. Based on the multi-dimensional damage feature vector, a classification model is trained using a support vector machine with cross-validation to form a diagnostic model that can identify damage to typical components.

[0070] Specifically, the degree of damage refers to the severity of damage to the core components of the power pole, which can usually be classified into different levels based on factors such as the area, depth, and degree of deformation of the damage, such as minor damage, moderate damage, and severe damage; annotation refers to manually or automatically marking the damage type and degree level on the point cloud samples of the core components of the power pole, providing supervision information for subsequent feature extraction and model training; multi-dimensional damage feature vectors refer to the set of vectors extracted from different angles and levels (such as geometry, texture, size changes, etc.) that can characterize the damage features of the core components.

[0071] Support vector machines (SVMs) are used for classification and regression analysis; here, they are used to build a classification model to identify typical component damage. Cross-validation is a method for evaluating model performance by dividing the dataset into multiple subsets and using them alternately for training and validation to improve the model's generalization ability and reliability. A classification model is a model that, after training, can classify input data (such as damage feature vectors) into predefined categories (such as different damage types and severity levels). A diagnostic model is a classification model used to diagnose damage to components of power poles.

[0072] In one feasible implementation, point cloud samples of core components of power poles with different degrees of damage are collected, such as point cloud samples containing damage types such as angle steel corrosion and steel pipe deformation; these samples are labeled to record the damage type (such as corrosion, deformation, etc.) and severity level (such as slight, moderate, severe) of each sample; multi-dimensional damage feature vectors are extracted from the labeled samples, such as extracting features of various dimensions such as geometric shape changes, surface texture features, and dimensional deviations, to construct multi-dimensional damage feature vectors.

[0073] Based on these multi-dimensional damage feature vectors, a support vector machine algorithm is adopted, and a classification model is trained through cross-validation. After training, a diagnostic model is formed that can identify typical component damage (such as missing bolts, angle steel deformation, steel pipe cracks, etc.). By labeling samples and extracting multi-dimensional feature vectors, and using support vector machines and cross-validation to train a high-precision diagnostic model, the automatic and accurate identification of power pole component damage can be achieved, improving the efficiency and intelligence level of fault diagnosis and ensuring the safe and stable operation of the power grid.

[0074] Furthermore, the method used in this application for verifying insulator deterioration faults on power poles includes: A subset of the point cloud of the insulator string is extracted to obtain the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt. Based on the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt, a spectral analysis operation is performed in conjunction with image data acquired by UAV vision to determine the insulator degradation level using a fuzzy evaluation method. The spectral analysis operation includes ultraviolet corona detection operation nodes, infrared thermal imaging analysis operation nodes, and visible light band spectral analysis operation nodes.

[0075] Specifically, an insulator string refers to a string-like structure composed of multiple insulators (components used to support and insulate power conductors), used in power poles to achieve electrical insulation and mechanical support functions; a point cloud subset refers to a partial set of data extracted from the entire point cloud dataset that is related to a specific target (such as an insulator string); a shed is a structure on the surface of an insulator used to enhance insulation performance and prevent flashover caused by pollution; insulators typically have multiple sheds; the surface area change rate refers to the proportion of change in the surface area of ​​the insulator sheds relative to its initial state due to pollution, aging, etc., and can be used to assess the degree of insulator degradation.

[0076] Surface roughness index is an indicator that quantifies the surface roughness of insulator skirts. Increased surface roughness may mean a decrease in insulation performance. Adjacent skirt spacing deviation rate refers to the degree of deviation between the actual spacing between adjacent skirts and the design standard spacing, which affects the electrical performance and mechanical stability of the insulator. Spectral analysis is a technique that obtains physical or chemical information by analyzing the spectral characteristics of an object's reflection or emission, and is used to detect the deterioration of insulators.

[0077] Ultraviolet corona detection uses ultraviolet sensors to detect corona discharge phenomena around insulators. Corona discharge may indicate local defects or performance degradation in the insulator. Infrared thermal imaging analysis uses an infrared thermal imager to detect the temperature distribution on the insulator surface. Temperature anomalies may suggest internal defects or aging. Visible light spectral analysis uses spectral information in the visible light band to assess the surface condition of the insulator, such as color changes and degree of contamination. Fuzzy evaluation methods can handle multi-index evaluation problems with fuzziness and uncertainty. They are used to determine the degradation level of insulators by combining multiple indicators (such as surface area change rate and surface roughness index) and using fuzzy membership functions and weight allocation to classify the degree of insulator degradation.

[0078] In one feasible implementation, a subset of the point cloud of the insulator string is extracted from the collected point cloud data of the power pole tower. For example, the point cloud data of the insulator string is identified and separated by a specific algorithm. For each skirt of each insulator string, the surface area change rate, surface roughness index and the deviation rate of the spacing between adjacent skirts are measured. Spectral analysis is performed in conjunction with the image data acquired by the UAV, specifically including ultraviolet corona detection, infrared thermal imaging analysis and visible light spectrum analysis.

[0079] By using the fuzzy evaluation method to determine the degradation level of insulators, and by extracting multi-dimensional features of insulator strings and combining them with spectral analysis, the degradation degree of insulators can be comprehensively and meticulously assessed. This provides key insulator status information for the entire fault diagnosis system, ensuring the timely detection of potential insulator fault risks, avoiding power outages that may be caused by insulator faults, ensuring the safe and stable operation of the power system, and effectively improving the reliability of the power grid.

[0080] In summary, the beneficial effects of the embodiments of this application are: By using point cloud data collected from power poles and associating it with the core components of the power poles to determine local geometric features (including the angle steel and steel pipes corresponding to the tower body, and the crossarms and grounding frames corresponding to the tower head), and by associating the point cloud data with the structural components of the power poles to determine global geometric features (including the tower body, tower head, and connecting rods), and by combining the local and global geometric features with image data collected by UAV vision and using voxel mesh downsampling to obtain fused point cloud data, and by comparing and analyzing the fused point cloud data with the standard geometric features in the data storage unit to confirm feature similarity and deviation values, the power poles are used to perform tilt fault verification, component damage fault verification, and insulator deterioration fault verification. A high-dimensional fault characteristic tensor is configured, and fault failure alerts are provided by combining the physical characteristics of the connection method under the constraint of connecting rods. This application provides a fault diagnosis method and system for power pole point cloud data. It achieves the separation of local geometric features of core components from global geometric features of architectural components, and combines UAV visual images and point cloud data to make up for the deficiencies of a single data source in terms of detailed representation and spatial structure. It establishes standard geometric features associated with pole model and parameters, and performs tilt fault verification, component damage fault verification, and insulator deterioration fault verification on power poles, thus ensuring the reliability of fault failure warnings.

[0081] Example 2, based on the same inventive concept as the fault diagnosis method for power pole point cloud data in the foregoing examples, such as... Figure 2 As shown in the figure, this application embodiment provides a fault diagnosis system for power pole point cloud data, wherein the system includes: The data acquisition module M100 is used to acquire point cloud data of power poles and towers, and associate it with the core components of the power poles and towers to determine local geometric features. The core components of the power poles and towers include the angle steel and steel pipes corresponding to the tower body, and the crossarms and grounding wire frame corresponding to the tower head.

[0082] The feature extraction module M200 is used to determine the global geometric features associated with the power pole structure components, including the tower body, tower head, and connecting rods.

[0083] The feature combination module M300 is used to combine the local geometric features and global geometric features based on the image data acquired by the UAV vision, and obtain fused point cloud data by voxel grid downsampling.

[0084] The fault verification module M400 is used to compare and analyze the fused point cloud data with the standard geometric features in the data storage unit to confirm the feature similarity and deviation value, and to perform tilt fault verification, component damage fault verification, and insulator deterioration fault verification on the power pole. It also configures a high-dimensional fault characteristic tensor and provides failure alerts based on the physical characteristics of the connection method under the limitation of the connecting rod.

[0085] Furthermore, the fault verification module M400 is also used to perform the following method: Based on the high-dimensional fault characteristic tensor, the fault failure components and fault evolution index are determined. According to the fault failure components and fault evolution index, the feasible region of the fault verification database is updated. The feasible region is mapped to the searched set of fault feature vectors, and the fault verification database is mapped to the tilt fault verification interval, component damage fault verification interval, and insulator deterioration fault verification interval. Simultaneously, using the weighted entropy value of the fault evolution index and diagnostic accuracy as the reward function, a reinforcement learning framework is constructed to dynamically optimize the diagnostic strategy. When the confidence level of the detected feature vector set is lower than a preset threshold, a contrastive learning unit is triggered to enhance the fault feature discrimination.

[0086] Furthermore, the fault verification module M400 is used to perform the following method: Determine the theoretical center axis of the power pole body and the theoretical center axis of the tower head; assess the degree of tilt of the power pole by measuring the angle and offset distance between the actual center axis and the theoretical center axis of the tower body and the theoretical center axis of the tower head; compare the degree of tilt of the power pole with a preset tilt threshold to confirm the tilt fault verification result.

[0087] Furthermore, the data acquisition module M100 is also used to perform the following methods: Based on the core components of the power pole, the pole point cloud data is associated and segmented to obtain multiple segmented point cloud data sets. This includes setting a density threshold and a neighborhood radius, traversing the pole point cloud data, collecting the number of points in the neighborhood of each point, and if the point density is greater than the density threshold, it is recorded as a core point; classifying the core point and the remaining points within its neighborhood radius into the same category to obtain multiple segmented point cloud data sets associated with the core components of the power pole; formulating a covariance matrix based on the multiple segmented point cloud data sets; obtaining the eigenvalues ​​and eigenvectors of the covariance matrix, sorting them according to the eigenvalues ​​from largest to smallest, selecting the eigenvectors corresponding to the first N eigenvalues, and setting an eigenvector matrix; and determining the local geometric features based on the eigenvector matrix.

[0088] Furthermore, the feature combination module M300 is also used to perform the following method: Point cloud data of similar power poles under different working conditions are introduced, and similar fused point cloud data corresponding to the core components and structural components of similar power poles are extracted. Based on the similar fused point cloud data, the corrosion coordinates of angle steel and the deformation coordinates of steel pipe are identified, and the corrosion depth and deformation parameters are extracted. The corrosion depth and deformation parameters are input into a pre-trained fatigue life prediction model to dynamically evaluate the remaining service life of the core components of the power pole.

[0089] Furthermore, the fault verification module M400 is also used to perform the following method: A standard feature vector library is established based on the standard geometric features; the standard feature vector library is stored in the data storage unit; based on the fused point cloud data of the same type, the standard feature vectors are associated and combined with the corresponding tower type and parameter information to obtain standard geometric features.

[0090] Furthermore, the fault verification module M400 is also used to perform the following method: According to the standard feature vector library, multiple feature vectors to be compared are identified, and the cosine similarity and Euclidean distance deviation value between the first feature vector to be compared and the associated standard feature vector are obtained. The first feature vector to be compared is any one of the multiple feature vectors to be compared. If the cosine similarity is lower than the preset similarity threshold or the Euclidean distance deviation value exceeds the dynamic compensation parameter, an anomaly alert is triggered, and the fault component type is located based on the deviation direction.

[0091] Furthermore, the fault verification module M400 is also used to perform the following method: Based on point cloud samples of core components of power poles with different degrees of damage, the damage type and severity level are labeled, and a multi-dimensional damage feature vector is constructed. Based on the multi-dimensional damage feature vector, a classification model is trained using a support vector machine with cross-validation to form a diagnostic model that can identify damage to typical components.

[0092] Furthermore, the fault verification module M400 is also used to perform the following method: A subset of the point cloud of the insulator string is extracted to obtain the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt. Based on the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt, a spectral analysis operation is performed in conjunction with image data acquired by UAV vision to determine the insulator degradation level using a fuzzy evaluation method. The spectral analysis operation includes ultraviolet corona detection operation nodes, infrared thermal imaging analysis operation nodes, and visible light band spectral analysis operation nodes.

[0093] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0094] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. A method for fault diagnosis of point cloud data of power poles, characterized in that, The method includes: Point cloud data of power poles is collected and associated with the core components of the power poles to determine local geometric features. The core components of the power poles include the angle steel and steel pipes corresponding to the tower body, and the crossarms and grounding wire frame corresponding to the tower head. Determine the global geometric features of the power pole structure components, including the tower body, tower head, and connecting members; Based on the image data acquired by the UAV, the local geometric features and global geometric features are combined, and voxel grid downsampling is used to obtain fused point cloud data; The fused point cloud data is compared and analyzed with the standard geometric features in the data storage unit to confirm the feature similarity and deviation values. The power pole is then checked for tilting faults, component damage faults, and insulator deterioration faults. A high-dimensional fault characteristic tensor is configured, and failure alerts are given in combination with the physical characteristics of the connection method under the limitation of the connecting rod.

2. The fault diagnosis method for power pole point cloud data as described in claim 1, characterized in that, The method for verifying tilting faults, component damage faults, and insulator deterioration faults of power poles, and configuring a high-dimensional fault characteristic tensor, includes: Based on the high-dimensional fault characteristic tensor, the faulty components and fault evolution index are determined. Based on the faulty components and the fault evolution index, the feasible region of the fault verification database is updated. The feasible region is mapped to the set of fault feature vectors obtained by the search. The fault verification database is mapped to the tilt fault verification interval, the component damage fault verification interval, and the insulator deterioration fault verification interval. Meanwhile, a reinforcement learning framework is constructed to dynamically optimize the diagnostic strategy by using the weighted entropy value of the fault evolution index and the diagnostic accuracy as the reward function. When the confidence of the detected feature vector set is lower than the preset threshold, the contrastive learning unit is triggered to enhance the distinguishability of fault features.

3. The fault diagnosis method for power pole point cloud data as described in claim 2, characterized in that, The method for checking tilt faults in power poles includes: Determine the theoretical center axis of the tower body and the theoretical center axis of the tower head of the power pole; The degree of tilt of the power pole is assessed by measuring the angle and offset distance between the actual center axis and the theoretical center axis of the tower body and the theoretical center axis of the tower head. The tilt degree of the power pole is compared with a preset tilt threshold to confirm the tilt fault verification result.

4. The fault diagnosis method for power pole point cloud data as described in claim 3, characterized in that, The method further includes collecting point cloud data of power poles and associating it with the core components of the power poles to determine local geometric features; and collecting such data. Based on the core components of the power pole, the pole point cloud data is associated and segmented to obtain multiple segmented point cloud data sets, including setting a density threshold and a neighborhood radius, traversing the pole point cloud data, collecting the number of points in the neighborhood of each point, and if the point density is greater than the density threshold, it is recorded as a core point. The core point and the remaining points within its neighborhood radius are classified into the same category to obtain a set of multiple segmented point cloud data associated with the core components of the power pole; The covariance matrix is ​​determined based on multiple segmented point cloud data sets; Obtain the eigenvalues ​​and eigenvectors of the covariance matrix, sort the eigenvalues ​​from largest to smallest, select the eigenvectors corresponding to the first N eigenvalues, and set the eigenvector matrix; The local geometric features are determined based on the eigenvector matrix.

5. The fault diagnosis method for power pole point cloud data as described in claim 4, characterized in that, The method further includes combining the local and global geometric features and using voxel grid downsampling to obtain fused point cloud data. Introduce point cloud data of similar power poles under different working conditions, and extract similar fused point cloud data corresponding to the core components and structural components of similar power poles; Based on the aforementioned fused point cloud data, the corrosion coordinates of angle steel and the deformation coordinates of steel pipe are identified, and the corrosion depth and deformation parameters are extracted. The corrosion depth and deformation parameters are input into a pre-trained fatigue life prediction model to dynamically assess the remaining service life of the core components of the power pole.

6. The fault diagnosis method for power pole point cloud data as described in claim 5, characterized in that, The method involves comparing and analyzing the fused point cloud data with standard geometric features in the data storage unit to confirm feature similarity and deviation values. A standard feature vector library is established based on the aforementioned standard geometric features; the standard feature vector library is stored in the aforementioned data storage unit; Based on the aforementioned fused point cloud data, the standard feature vector is associated and combined with the corresponding tower model and parameter information to obtain the standard geometric features.

7. The fault diagnosis method for power pole point cloud data as described in claim 6, characterized in that, According to the standard feature vector library, multiple feature vectors to be compared are identified, and the cosine similarity and Euclidean distance deviation between the first feature vector to be compared and the associated standard feature vector are obtained, wherein the first feature vector to be compared is any one of the multiple feature vectors to be compared. If the cosine similarity is lower than the preset similarity threshold or the Euclidean distance deviation exceeds the dynamic compensation parameter, an anomaly alert is triggered, and the faulty component type is located based on the deviation direction.

8. The fault diagnosis method for power pole point cloud data as described in claim 2, characterized in that, The method for verifying component faults on power poles includes: Based on point cloud samples of core components of power poles with different degrees of damage, the damage type and degree level are labeled, and a multi-dimensional damage feature vector is constructed. Based on the aforementioned multi-dimensional damage feature vectors, a support vector machine is used to train a classification model using cross-validation, forming a diagnostic model that can identify damage to typical components.

9. The fault diagnosis method for power pole point cloud data as described in claim 2, characterized in that, The method for verifying insulator deterioration faults on power poles includes: Extract a subset of the point cloud of the insulator string to obtain the surface area change rate, surface roughness index and adjacent skirt spacing deviation rate of each skirt of the insulator string; Based on the surface area change rate, surface roughness index, and adjacent skirt spacing deviation rate of each skirt of the insulator string, a spectral analysis operation is performed using image data acquired by UAV vision, and the insulator degradation level is determined by fuzzy evaluation method. The spectral analysis operation includes an ultraviolet corona detection operation node, an infrared thermal imaging analysis operation node, and a visible light band spectral analysis operation node.

10. A fault diagnosis system for point cloud data of power poles, characterized in that, A system for implementing a fault diagnosis method for power pole point cloud data according to any one of claims 1-9, the system comprising: The data acquisition module is used to collect point cloud data of power poles and towers, and associate it with the core components of the power poles and towers to determine local geometric features. The core components of the power poles and towers include angle steel and steel pipes corresponding to the tower body, and crossarms and grounding frames corresponding to the tower head. The feature extraction module is used to determine the global geometric features associated with the power pole structure components, including the tower body, tower head, and connecting rods. The feature combination module is used to combine the local geometric features and global geometric features based on the image data acquired by the UAV vision, and obtain fused point cloud data by voxel grid downsampling; The fault verification module is used to compare and analyze the fused point cloud data with the standard geometric features in the data storage unit to confirm the feature similarity and deviation value, and to perform tilt fault verification, component damage fault verification, and insulator deterioration fault verification on the power pole. It also configures a high-dimensional fault characteristic tensor and provides failure alerts based on the physical characteristics of the connection method under the limitation of the connecting rod.

Images