Defect detection method for unmanned aerial vehicle inspection equipment of transformer substation

Abstract

The invention discloses a defect detection method for transformer substation unmanned aerial vehicle inspection equipment, and belongs to the technical field of transformer substation defect detection. Performing temperature field inversion correction on the monitoring data, establishing an incidence matrix according to sub-pixel displacement quantification mechanical deformation, and matching a preset defect feature library to obtain a first detection result; according to the first detection result and an assembly digital twin model, correcting the incidence matrix and positioning an abnormal area; inputting the corrected incidence matrix into a defect detection model embedded with the component prior constraint, and outputting a second detection result; and constructing a causal graph of the inspection records and the assembly working conditions, and outputting a predictive maintenance strategy according to the second detection result and the abnormal region. According to the invention, the problems of low detection precision and insufficient association degree with the component state in the aspect of component defect detection in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of substation defect detection, and particularly to a defect detection method for unmanned aerial vehicle (UAV) inspection equipment for substations. Background Art

[0002] In recent years, with the accelerated advancement of the construction of smart grids, substations, as the core hubs of power systems, the real-time monitoring and maintenance of the operating status of their equipment have become particularly important. The traditional manual inspection method is not only inefficient but also difficult to cover the high-altitude and hidden parts of the equipment, making it difficult to meet the high requirements of modern power grids for safety and reliability. In this context, UAV inspection technology has gradually become an important means for substation equipment inspection due to its high efficiency, flexibility, and safety.

[0003] Although significant progress has been made in the defect detection of substation components using UAV inspection technology, there are still problems with low detection accuracy and insufficient correlation with component status in the existing technology. On the one hand, the detection accuracy is limited by various factors. First, during flight, the UAV may be affected by environmental factors such as wind and light, resulting in problems such as blurred and jittery images or videos, which in turn affect the accuracy of defect recognition. Second, complex textures, reflections, and obstructions on the surface of the equipment may also interfere with defect detection, making it difficult to accurately identify some subtle defects. In addition, existing intelligent recognition algorithms still have misjudgment and missed judgment situations when dealing with complex scenarios and diverse defect types, and need to be further optimized and improved. On the other hand, there is a lack of correlation with component status in the existing technology. Currently, most UAV inspection systems mainly focus on visual defect detection on the surface of the equipment, lacking consideration of comprehensive information such as the internal status and operating parameters of the equipment. This results in the detection results often only reflecting the abnormal conditions on the surface of the equipment, and it is difficult to deeply reveal potential faults or performance degradation trends inside the equipment. Therefore, when formulating maintenance plans, it may be difficult for staff to accurately assess the actual status and remaining life of the equipment, thus affecting the scientificity and effectiveness of maintenance decisions. Summary of the Invention

[0004] The technical problem solved by the present invention is the problems of low detection accuracy and insufficient correlation with component status existing in the existing technology in component defect detection.

[0005] To solve the above technical problems, the present invention provides the following technical solutions:

[0006] The present invention provides a defect detection method for unmanned aerial vehicle inspection equipment for substations, including:

[0007] Collecting the monitoring data of the UAV;

[0008] Perform temperature field inversion correction on the monitoring data, establish a correlation matrix based on sub-pixel displacement quantization of mechanical deformation, and match the preset defect feature library to obtain the first detection result;

[0009] Modify the correlation matrix according to the first detection result and the component digital twin model, and locate the abnormal area;

[0010] Input the modified correlation matrix into the defect detection model embedded with prior constraints of the component, and output the second detection result;

[0011] Construct a causal graph of the inspection record and the component working condition, and output the predictive maintenance strategy according to the second detection result and the abnormal area.

[0012] Furthermore, the monitoring data includes visible light images, infrared images, and LiDAR point cloud data; among them, the method for collecting the monitoring data of the drone includes:

[0013] Control the drone equipped with a multi-spectral sensor, an infrared thermal imager, and a LiDAR sensor to inspect the substation according to a preset route;

[0014] Synchronize the acquisition clocks of the multi-spectral sensor, the infrared thermal imager, and the LiDAR sensor using the GPS PPS pulse signal;

[0015] In an environment with electromagnetic interference, adopt the adaptive frequency hopping communication technology to switch the 2.4G / 5.8G / LoRa frequency bands for data transmission, and collect the visible light images, infrared images, and LiDAR point cloud data of the components;

[0016] Introduce an adaptive weight mechanism into the point-to-point distance error function to improve the iterative closest point ICP algorithm;

[0017] Construct a three-dimensional coordinate system of the component based on the LiDAR point cloud data, and use the improved iterative closest point ICP algorithm to map the pixels of the visible light image and the infrared image to a unified spatial coordinate system to obtain the monitoring data of the drone.

[0018] Furthermore, constructing a three-dimensional coordinate system of the component based on the LiDAR point cloud data, and using the improved iterative closest point ICP algorithm to map the pixels of the visible light image and the infrared image to a unified spatial coordinate system to obtain the monitoring data of the drone includes:

[0019] Based on the LiDAR point cloud data, calculate the geometric center of the component as the coordinate origin, use the principal component analysis PCA to fit the surface normal vector of the component as the coordinate Z-axis, align the coordinate X-axis to the due north direction of the substation building direction, and establish a three-dimensional coordinate system of the component;

[0020] Use the improved Iterative Closest Point (ICP) algorithm to calculate the transformation matrix between the visible light image and the infrared image in the three-dimensional coordinate system of the component;

[0021] Adopt the Scale-Invariant Feature Transform (SIFT) algorithm to extract the scale-invariant feature points of the visible light image and the infrared image, and perform preliminary matching through a Fast Library for Approximate Nearest Neighbors (FLANN) matcher to obtain the matching point pairs;

[0022] Generate a depth map using LiDAR point cloud data, and eliminate the mismatched points in the matching point pairs by verifying the geometric consistency of the component to obtain the corrected matching point pairs;

[0023] Calculate the centroid coordinates of the corrected matching points, and perform singular value decomposition on the covariance matrix of the centroid coordinates to solve the optimal values of the rotation matrix and the translation vector;

[0024] Optimize the transformation matrix using the optimal values of the rotation matrix and the translation vector, and map the pixels of the visible light image and the infrared image to a unified spatial coordinate system through the inverse projection formula to obtain the monitoring data of the unmanned aerial vehicle (UAV);

[0025] Furthermore, perform temperature field inversion correction on the monitoring data, establish a correlation matrix according to the sub-pixel displacement quantization of mechanical deformation, and match the pre-set defect feature library to obtain the first detection result, including:

[0026] Measure the reflectivity and emissivity of the UAV material at different temperatures to pre-construct the reflectivity-emissivity curve;

[0027] Based on the UAV material and the reflectivity-emissivity curve, obtain the reflectivity and emissivity data of the UAV material;

[0028] According to the infrared image, adopt the inverse Planck function method to preliminarily invert the temperature field of the monitoring area, and convert the infrared radiation intensity into the temperature distribution characteristics using the obtained reflectivity and emissivity data of the UAV material to obtain the preliminary temperature field inversion result;

[0029] Establish a correction model based on the influence of the environmental parameters of the monitoring area and the flight attitude of the UAV on the temperature field inversion, and correct the preliminary inversion result to obtain the temperature field inversion result;

[0030] Register the visible light image and the infrared thermal imaging in the monitoring data of the UAV collected at different temperatures according to the temperature field inversion result, and calculate the sub-pixel displacement between the images;

[0031] Establish a mathematical model between the mechanical deformation and the sub-pixel displacement according to the sub-pixel displacement between the images and the UAV material, and convert the sub-pixel displacement between the images into mechanical deformation characteristics;

[0032] Construct a correlation matrix between the temperature distribution feature and the mechanical deformation feature, match the correlation matrix with a preset defect feature library, and obtain a first detection result;

[0033] The preset defect feature library stores the defect types of the equipment corresponding to the correlation matrix between the temperature distribution feature and the mechanical deformation feature;

[0034] The first detection result includes the component defect type.

[0035] Further, the method for constructing the component digital twin model includes:

[0036] Based on the LiDAR point cloud data, use the B-Rep and CSG hybrid modeling technology to construct a component digital twin model with LOD4-level accuracy;

[0037] Preset a fault model in the component digital twin model, and the fault model is used to simulate the behavior characteristics of the component under different component defect types;

[0038] Use the Kalman filter to dynamically synchronize the behavior characteristics of the component to the component digital twin model according to the monitoring data of the unmanned aerial vehicle.

[0039] Further, according to the first detection result and the component digital twin model, modify the parameters of the correlation matrix and locate the abnormal area, including:

[0040] Map the first detection result to the fault model of the component digital twin model, simulate the behavior characteristics of the component under the first detection result, and obtain the actual monitoring data of the component digital twin model;

[0041] Calculate the temperature distribution feature deviation and the mechanical deformation feature deviation between the actual monitoring data and the expected monitoring data mapped by the first detection result to the component digital twin model, and obtain the result of the difference analysis;

[0042] According to the result of the difference analysis, modify the parameters of the correlation matrix;

[0043] The abnormal area includes a temperature abnormal area and a deformation abnormal area;

[0044] If the temperature distribution feature deviation exceeds the preset temperature threshold, the area where the temperature distribution feature deviation occurs is used as the temperature abnormal area;

[0045] If the mechanical deformation feature deviation exceeds the preset temperature threshold, the area where the temperature distribution feature deviation occurs is used as the deformation abnormal area.

[0046] Further, the component prior constraints include: the temperature constraint of the component, the geometric shape constraint of the component, the position constraint of the component, the material constraint of the component, and the installation accuracy constraint of the component.

[0047] Furthermore, the defect detection model with prior constraints on the embedding component includes:

[0048] A feature extraction module, including a temperature extraction branch and a deformation extraction branch, is used to extract the temperature distribution features and mechanical deformation features of the corrected correlation matrix and input them into the prior constraint embedding layer;

[0049] The prior constraint embedding layer includes a temperature constraint function of the component, a geometric shape constraint function of the component, a position constraint function of the component, a material constraint function of the component, and an installation accuracy function of the component, and is used to screen the temperature distribution features and mechanical deformation features and input them into the dynamic sparse inference module;

[0050] The dynamic sparse inference module is used to select the temperature distribution features and mechanical deformation features whose relevance to defect detection exceeds the threshold and input them into the defect detection and classification module;

[0051] The defect detection and classification module is used for the YOLOv8 network to analyze according to the temperature distribution features and mechanical deformation features and output the second detection result;

[0052] The second detection result includes the defect type, defect location, defect severity, and confidence level of the component.

[0053] Furthermore, a causal graph of the inspection record and the component working condition is constructed, including:

[0054] According to the inspection record and the component working condition, an inspection record node and a component working condition node are established;

[0055] Based on the K2 algorithm with the maximum posterior probability as the optimization goal, heuristic search is performed on the inspection node and the component working condition node according to domain knowledge to establish a directed edge, and a directed acyclic causal graph of the inspection record and the component working condition including the causal relationship strength weight is output;

[0056] The directed edge is used to represent the causal relationship between the inspection node and the component working condition node;

[0057] The length of the directed edge is used to represent the causal strength between the inspection node and the component working condition node;

[0058] The inspection node includes the observation indicators in the inspection record, including temperature distribution and mechanical deformation amount;

[0059] The component working condition node includes the state variables of the component working condition, including normal operation, minor fault, moderate fault, and severe fault.

[0060] Furthermore, a predictive maintenance strategy is output according to the second detection result and the abnormal area, including:

[0061] The second detection result includes the defect type, defect location, defect severity, and confidence level of the component;

[0062] The abnormal area includes the shape and size of the abnormal part in the component;

[0063] Risk weights are assigned to the defect type, defect location, defect severity, confidence level of the component, and the shape and size of the abnormal part in the component respectively, and the risk score of the defect is calculated;

[0064] According to the risk score of the defect, the risk level is divided and a predictive maintenance strategy is formulated.

[0065] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0066] 1. By obtaining the monitoring data of the unmanned aerial vehicle (UAV), combining temperature field inversion correction, component mechanical deformation quantification analysis, and digital twin model dynamic correction technology, the present invention constructs a defect detection method for UAV inspection equipment for substations. It not only realizes the accurate positioning and classification of defects, but also can output a predictive maintenance strategy through causal graph reasoning of inspection records and component working conditions, innovatively upgrading traditional after-the-fact maintenance to proactive maintenance based on real-time status data, significantly improving the operation and maintenance efficiency of substations, reducing the risk of sudden failures, and solving the problems of low detection accuracy and insufficient correlation with component status existing in the prior art in component defect detection.

[0067] 2. The present invention innovatively couples the physical properties of materials with image analysis in depth, forms a two-way feedback through temperature field inversion correction and mechanical deformation quantification, breaks through the limitations of single-modal detection, and also constructs a reflectivity-emissivity curve library of UAV materials in advance, combines the Planck inverse function method to realize temperature field inversion correction, quantifies mechanical deformation through sub-pixel displacement, establishes a correlation matrix between temperature distribution and mechanical deformation, and matches the preset defect feature library, thereby improving the accuracy of component defect recognition.

[0068] 3. The present invention introduces component prior constraints into the defect detection model, automatically screens highly relevant features through a dynamic sparse inference module, and cooperates with the YOLOv8 network to achieve end-to-end defect detection and classification. Compared with the unconstrained model, it improves the false alarm rate and inference speed of defects, and at the same time quantifies the defect severity through the confidence level, providing more refined gradient information for operation and maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic diagram of the basic process of a defect detection method for UAV inspection equipment for substations provided by an embodiment of the present invention;

[0070] Figure 2Schematic table of the reflectivity and emissivity of the UAV material provided by an embodiment of the present invention;

[0071] Figure 3 Structural schematic diagram of a defect detection model with prior constraints for embedded components provided by an embodiment of the present invention. Detailed implementation manners

[0072] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments.

[0073] Embodiment 1

[0074] As Figure 1 shown, this embodiment introduces a defect detection method for substation UAV inspection equipment, including:

[0075] Step 1: Collect monitoring data of the UAV.

[0076] The present invention obtains the original state information of substation components through the monitoring data of the UAV, providing a data basis for subsequent analysis and processing. The UAV has the characteristics of flexibility and high efficiency, and can quickly and comprehensively collect data such as images and thermal images of each component in the substation, overcoming the problems of low efficiency and limited scope of traditional manual inspections, ensuring that sufficient rich and accurate equipment state information can be obtained, and providing comprehensive data support for subsequent defect detection.

[0077] Step 2: Perform temperature field inversion correction on the monitoring data, establish a correlation matrix based on sub-pixel displacement quantization of mechanical deformation, and match the preset defect feature library to obtain the first detection result.

[0078] The present invention establishes a correlation matrix through the temperature distribution and mechanical deformation, which can comprehensively consider the thermal and mechanical states of the components and break through the limitations of single-modal detection. Different defects may lead to different change patterns of temperature and mechanical deformation. By establishing a correlation matrix, the state of the components can be analyzed more comprehensively. Matching the correlation matrix with the preset defect feature library can quickly identify common defect types and obtain the first detection result. The preset defect feature library contains the feature patterns of various known defects. By matching with the correlation matrix, it can be preliminarily judged whether there are defects in the components and the types of defects.

[0079] Step 3: Correct the correlation matrix according to the first detection result and the component digital twin model, and locate the abnormal area.

[0080] The present invention combines the first detection result with the component digital twin model to correct the correlation matrix. The digital twin model is a virtual representation of the component, containing information such as the geometric structure and physical properties of the component. By comparing and analyzing with the digital twin model, the correlation matrix can be further optimized to improve its ability to reflect the actual state of the component. Based on the corrected correlation matrix, the abnormal area in the component can be more accurately located. The abnormal area is the location where the defect is located. By locating the abnormal area, a clear target can be provided for subsequent detailed detection and maintenance.

[0081] Step 4: Input the corrected correlation matrix into the defect detection model embedded with the prior constraints of the component, and output the second detection result.

[0082] The present invention embeds the prior constraints of the component in the defect detection model, which can utilize the prior knowledge such as the physical properties and structural characteristics of the component to improve the detection accuracy of the model. The prior constraints can help the defect detection model better understand the difference between the normal state and the abnormal state of the component, reducing false alarms and missed detections. The present invention also performs end-to-end defect detection and classification through the YOLOv8 network to achieve rapid detection and classification of component defects and output the second detection result.

[0083] Step 5: Construct a causal graph of the inspection record and the component working condition, and output the predictive maintenance strategy according to the second detection result and the abnormal area.

[0084] The present invention constructs a causal graph of the inspection record and the component working condition to analyze the causal relationship between various factors in the inspection record and the component working condition. Through the causal graph, the present invention can deeply understand the causes and development processes of component defects, providing a theoretical basis for predictive maintenance. According to the second detection result and the abnormal area, combined with the analysis results of the causal graph, the predictive maintenance strategy is output.

[0085] The predictive maintenance strategy can formulate a maintenance plan in advance according to the actual state and potential risks of the component, avoid the occurrence of sudden failures, reduce the operation and maintenance costs, and improve the reliability and safety of the substation. At the same time, upgrading the traditional after-fact maintenance to proactive maintenance based on real-time status data significantly improves the operation and maintenance efficiency of the substation.

[0086] Embodiment 2

[0087] With the same inventive concept as Embodiment 1, this embodiment introduces the specific steps of a defect detection method for substation UAV inspection equipment, including:

[0088] Step 1: Collect the monitoring data of the UAV.

[0089] In this embodiment, the monitoring data includes visible light images, infrared images, and LiDAR point cloud data. Among them, the method for collecting the monitoring data of the UAV includes:

[0090] Step 1.1: Control the UAV equipped with a multispectral sensor, an infrared thermal imager, and a LiDAR sensor to patrol the substation according to a preset route.

[0091] The multispectral sensor can capture spectral information in different bands, which can reflect the material characteristics, color changes, etc. of the component surface; the infrared thermal imager can detect the temperature distribution on the component surface to discover potential overheating defects; the LiDAR sensor can obtain the three-dimensional point cloud data of the component to accurately depict the geometric shape and spatial position of the component. By carrying these three sensors, the UAV can obtain detailed information about the substation components from multiple dimensions, providing a rich data basis for subsequent defect detection.

[0092] Patrolling according to the preset route can ensure that the UAV comprehensively and orderly covers each component in the substation, avoiding omission of important areas. The preset route can be optimized according to the layout of the substation, the importance and distribution of the components, improving the patrol efficiency and reducing the patrol time and labor costs.

[0093] Step ၁.၂: Synchronize the acquisition clocks of the multispectral sensor, the infrared thermal imager, and the LiDAR sensor using the GPS PPS pulse signal.

[0094] When different types of sensors collect data, if the clocks are not synchronized, it will cause deviations in the collected data in the time dimension, affecting subsequent data fusion and analysis. Using the GPS PPS (Pulse Per Second) pulse signal to synchronize the acquisition clocks can ensure that the multispectral sensor, the infrared thermal imager, and the LiDAR sensor start collecting data at the same moment, ensuring the time consistency of the data.

[0095] The synchronously collected data can more accurately reflect the state information of the component at the same moment. When performing data fusion and analysis, it can avoid misjudgment and missed judgment caused by time deviation, improving the accuracy of defect detection.

[0096] Step 1.3: In an environment with electromagnetic interference, adopt adaptive frequency hopping communication technology to switch the 2.4G / 5.8G / LoRa frequency bands for data transmission, and collect the visible light images, infrared images, and LiDAR point cloud data of the components.

[0097] There are various electromagnetic interference sources in the substation, such as high-voltage equipment, communication equipment, etc., which will affect the normal transmission of data. By adopting the adaptive frequency hopping communication technology, it can automatically switch between the 2.4G / 5.8G / LoRa frequency bands according to the change of the electromagnetic environment, avoid the frequency bands with severe interference, and ensure the stability and reliability of data transmission.

[0098] At the same time, collecting the visible light image, infrared image and LiDAR point cloud data of the component can comprehensively reflect the state information of the component from different angles. The visible light image can intuitively display the appearance characteristics of the component; the infrared image can detect the temperature abnormality of the component; the LiDAR point cloud data can accurately measure the three-dimensional shape and size of the component. The comprehensive collection of multi-type data provides richer information for subsequent defect detection and analysis.

[0099] Different frequency bands have different characteristics and application scenarios. The 2.4G frequency band has a relatively long transmission distance, but relatively large interference; the 5.8G frequency band has a high transmission rate, but a short transmission distance; the LoRa frequency band has the advantages of low power consumption and long-distance transmission. By adaptively switching the frequency band, it can select the most suitable frequency band for data transmission according to the actual environment of the substation and the data transmission requirements, and improve the efficiency and reliability of data transmission.

[0100] Step 1.4: Introduce an adaptive weight mechanism into the point-to-point distance error function to improve the iterative closest point (ICP) algorithm.

[0101] The iterative closest point (ICP) algorithm is often used for the registration of point cloud data. However, in practical applications, due to factors such as noise and occlusion of point cloud data, the traditional ICP algorithm will have problems with low registration accuracy. Introducing an adaptive weight mechanism into the point-to-point distance error function can assign different weights to each point in the point cloud data according to the reliability and importance of each point.

[0102] For points with high reliability and great influence on the registration result, larger weights are assigned; for noise points or points with low reliability, smaller weights are assigned. This can improve the registration accuracy and make the registered point cloud data more accurately reflect the actual shape and position of the component.

[0103] The adaptive weight mechanism can make the ICP algorithm more robust to noise and outliers in the point cloud data. In the presence of noise and outliers, the traditional ICP algorithm may be interfered, resulting in inaccurate registration results. By introducing adaptive weights, the influence of noise and outliers can be effectively suppressed, and the algorithm can maintain stable performance under different qualities of point cloud data.

[0104] In the defect detection of substation components, it may be necessary to fuse the point cloud data collected by different sensors. High-precision point cloud registration is a prerequisite for effective data fusion. The improved ICP algorithm can improve the accuracy of point cloud registration, thereby optimizing the effect of data fusion, enabling the fused data to more accurately reflect the state information of the components, and providing a more reliable basis for subsequent defect detection and analysis.

[0105] Step 1.5: Based on the LiDAR point cloud data, construct a three-dimensional coordinate system for the component, and use the improved iterative closest point (ICP) algorithm to map the pixels of the visible light image and the infrared image to a unified spatial coordinate system to obtain the monitoring data of the drone, including:

[0106] Step 1.5.1: Based on the LiDAR point cloud data, calculate the geometric center of the component as the coordinate origin, use principal component analysis (PCA) to fit the surface normal vector of the component as the coordinate Z-axis, and align the coordinate X-axis to the due north direction of the substation building direction to establish a three-dimensional coordinate system for the component.

[0107] In this embodiment, by establishing a three-dimensional coordinate system for the component, a unified spatial reference framework is constructed, with the geometric center of the component as the coordinate origin, making the description of the positions of each point on the component more intuitive and accurate. When determining the axis directions, using principal component analysis (PCA) to fit the surface normal vector of the component as the coordinate Z-axis can accurately reflect the main orientation and surface characteristics of the component, helping to deeply understand the attitude of the component in space. Aligning the coordinate X-axis to the due north direction of the substation building direction enables the three-dimensional coordinate system of the component to be associated with the overall layout of the substation, improving the accuracy of relative position analysis between multiple components and also enhancing the accuracy of interactive analysis between the component and the substation environment.

[0108] Step 1.5.2: Use the improved iterative closest point (ICP) algorithm to calculate the transformation matrix between the visible light image and the infrared image in the three-dimensional coordinate system of the component.

[0109] Due to differences in acquisition principles and perspectives, there may be deviations in space between the visible light image and the infrared image. Calculating the transformation matrix between the visible light image and the infrared image can align the two images to the same coordinate system, facilitating subsequent fusion analysis and defect detection.

[0110] The improved iterative closest point (ICP) algorithm has improved in registration accuracy and robustness, and can better handle noise and outliers in point cloud data. When calculating the transformation matrix between the visible light image and the infrared image, the improved iterative closest point (ICP) algorithm can more accurately find the corresponding relationship between the visible light image and the infrared image, thereby improving the registration accuracy.

[0111] After aligning the visible light image and the infrared image by calculating the transformation matrix, the advantages of both images are fully utilized for multimodal analysis. The visible light image provides rich color and texture information, while the infrared image reflects the temperature distribution of components. Combining the two can detect component defects more comprehensively.

[0112] Step 1.5.3: Use the SIFT algorithm to extract scale-invariant feature points from the visible light image and the infrared image, and perform preliminary matching through the FLANN matcher to obtain matching point pairs.

[0113] The SIFT (Scale-Invariant Feature Transform) algorithm can extract feature points with scale invariance in the image and describe them. The scale-invariant feature points are robust to image rotation, scaling, illumination changes, etc., and can accurately identify key features in the image under different conditions.

[0114] The FLANN (Fast Library for Approximate Nearest Neighbors) matcher is an efficient feature point matching algorithm that can quickly find similar feature point pairs in the feature point sets of two images. Through preliminary matching, a set of possible matching point pairs can be obtained, and the obtained matching point pairs are important inputs for subsequent steps such as verifying geometric consistency and calculating the transformation matrix. Accurate matching point pairs can improve the accuracy and reliability of subsequent processing.

[0115] Step 1.5.4: Generate a depth map using LiDAR point cloud data, and eliminate the mismatched points in the matching point pairs by verifying the geometric consistency of the components to obtain the corrected matching point pairs.

[0116] The depth map generated from LiDAR point cloud data can provide three-dimensional depth information of the component surface. Using the depth information, the geometric consistency of the matching point pairs can be verified to determine whether the matching point pairs conform to the actual geometric shape of the component. In the actual matching process, due to reasons such as image noise and feature point similarity, mismatched points will be generated. By verifying the geometric consistency, these mismatched points can be effectively eliminated, improving the accuracy of the matching point pairs. The corrected matching point pairs have higher quality and can provide a more reliable data basis for subsequent transformation matrix calculation and optimization, thereby improving the accuracy and reliability of the entire data processing process.

[0117] Step 1.5.5: Calculate the centroid coordinates of the corrected matching points, and perform singular value decomposition on the covariance matrix of the de-centered coordinates to solve the optimal values of the rotation matrix and the translation vector.

[0118] To calculate the centroid coordinates of the corrected matching points, the coordinates of the matching point pairs can be centralized to simplify the subsequent calculation process. The centroid coordinates reflect the overall position information of the matching point pairs. The covariance matrix of the centroid coordinates can describe the distribution relationship and correlation between the matching point pairs. By performing singular value decomposition on the covariance matrix, the main change directions and degrees of the matching point pairs can be extracted, providing a basis for solving the rotation matrix and translation vector. The results of the singular value decomposition can be used to solve the optimal values of the rotation matrix and translation vector, minimizing the error between the transformed matching point pairs. These optimal values can more accurately describe the spatial transformation relationship between the visible light image and the infrared image.

[0119] Step 1.5.6: Optimize the transformation matrix using the optimal values of the rotation matrix and translation vector, and map the pixels of the visible light image and the infrared image to a unified spatial coordinate system through the inverse projection formula to obtain the monitoring data of the unmanned aerial vehicle.

[0120] Optimizing the transformation matrix using the optimal values of the rotation matrix and translation vector can further improve the accuracy of the transformation matrix, making the registration between the visible light image and the infrared image more accurate. By mapping the pixels of the visible light image and the infrared image to a unified spatial coordinate system through the inverse projection formula, the fusion of the two images in space is achieved. Under the unified spatial coordinate system, the images can be conveniently analyzed and processed to extract more valuable information. The obtained monitoring data of the unmanned aerial vehicle contains the information of the visible light image and the infrared image in the unified spatial coordinate system, which can more comprehensively and accurately reflect the state of the components. These data provide rich information support for subsequent defect detection, analysis, and predictive maintenance, helping to improve the efficiency and reliability of substation operation and maintenance.

[0121] Step 2: Perform temperature field inversion correction on the monitoring data, establish a correlation matrix based on sub-pixel displacement quantization of mechanical deformation, and match the preset defect feature library to obtain the first detection result, including:

[0122] Step 2.1: Measure the reflectivity and emissivity of the unmanned aerial vehicle material at different temperatures to pre-construct a reflectivity-emissivity curve.

[0123] Measure the reflectivity and emissivity of the unmanned aerial vehicle material at different temperatures to pre-construct a reflectivity-emissivity curve. The schematic table of the measured reflectivity and emissivity of the unmanned aerial vehicle material is as Figure 2 shown.

[0124] The reflectivity and emissivity of drone materials will change at different temperatures. By measuring the reflectivity and emissivity of drone materials and constructing a reflectivity-emissivity curve, the optical properties of drone materials under different temperature conditions can be accurately described, providing a key material parameter basis for subsequent temperature field inversion based on infrared images, which helps to improve the accuracy of temperature inversion.

[0125] Step 2.2: Based on the drone material and the reflectivity-emissivity curve, obtain the reflectivity and emissivity data of the drone material.

[0126] Accurate reflectivity and emissivity data improves the accuracy of converting infrared radiation intensity to temperature distribution. Dynamically extracting the corresponding reflectivity and emissivity from the reflectivity-emissivity curve based on the actual drone material and the temperature conditions during monitoring ensures the use of the most appropriate parameters for temperature field inversion in different situations, improving the algorithm's flexibility and accuracy. Directly acquiring data through pre-built curves avoids complex real-time measurements and calculations, improving data processing efficiency while ensuring data reliability.

[0127] Step 2.3: Based on the infrared image, use the Planck inverse function method to perform a preliminary inversion of the temperature field of the monitored area. Use the acquired reflectivity and emissivity data of the drone material to convert the infrared radiation intensity into temperature distribution characteristics to obtain the preliminary inversion results of the temperature field.

[0128] The Planck inverse function method is a commonly used method for converting infrared radiation intensity to temperature. Combined with the reflectivity and emissivity data of drone materials, this method can convert the radiation information in infrared images into temperature distribution characteristics, initially constructing the temperature field of the monitored area. The temperature field is one of the key indicators of component operating status. The preliminary inversion results can intuitively display the temperature distribution on the component surface, helping to identify potential overheating defects such as poor contact and insulation degradation, and facilitate a more comprehensive analysis of component status.

[0129] Step 2.4: Establish a correction model based on the environmental parameters of the monitoring area and the influence of the UAV's flight posture on the temperature field inversion, correct the preliminary inversion results, and obtain the temperature field inversion results.

[0130] Environmental parameters of the monitoring area such as wind speed, humidity, etc. and the flight attitude of the UAV such as pitch angle, yaw angle, etc. will affect the measurement of infrared radiation and the inversion of the temperature field. Establishing a correction model can quantify and compensate for these effects, thereby improving the accuracy of the temperature field inversion. Through the correction model, the errors caused by environmental factors and flight attitude can be reduced, making the temperature field inversion result more accurate and reliable. This helps to improve the reliability and stability of the entire defect detection system. In actual substation inspections, the environmental conditions and flight attitude are constantly changing. The correction model can adapt to these changes and ensure accurate temperature field inversion results under different conditions.

[0131] Step 2.5: Register the visible light image and the infrared thermal image in the monitoring data of the UAV collected at different temperatures according to the temperature field inversion result, and calculate the sub-pixel displacement between the images.

[0132] The visible light image and the infrared thermal image contain different information of the component. Registration is the prerequisite for fusing and analyzing the visible light image and the infrared thermal image. By registering through the temperature field inversion result, it can ensure that the two images are spatially aligned. Calculating the sub-pixel displacement between the images can detect the tiny deformation and displacement changes on the surface of the component. Tiny deformation and displacement changes are early signs of potential defects in the component. These problems can be detected in a timely manner through sub-pixel displacement analysis. Sub-pixel level displacement calculation can improve the sensitivity to the change of the component state and provide a more accurate data basis for subsequent mechanical deformation feature extraction and analysis.

[0133] Step 2.6: Establish a mathematical model between the mechanical deformation and the sub-pixel displacement according to the sub-pixel displacement between the images and the UAV material, and convert the sub-pixel displacement between the images into mechanical deformation features.

[0134] By establishing a mathematical model to connect the sub-pixel displacement between the images with the mechanical deformation, the intuitive image displacement information can be converted into quantitative mechanical deformation features, which helps to more accurately evaluate the structural health status of the component. By considering the characteristics of the UAV material, the mathematical model is made more in line with the actual situation. Different materials will produce different deformation responses when stressed. The model established by combining the material characteristics can improve the accuracy of mechanical deformation feature extraction. Mechanical deformation features are important bases for judging whether there are structural defects in the component. Converting the sub-pixel displacement into mechanical deformation features can provide more effective information for subsequent defect diagnosis and classification.

[0135] ]>Step 2.7: Construct a correlation matrix between the temperature distribution features and the mechanical deformation features, and match the correlation matrix with the preset defect feature library to obtain the first detection result.

[0136] Constructing the correlation matrix between the temperature distribution characteristics and the mechanical deformation characteristics can comprehensively consider the thermal and mechanical states of the components. Different defects may lead to different change patterns of temperature and mechanical deformation. Through the correlation matrix, the state of the components can be comprehensively analyzed. Matching the correlation matrix with the pre-set defect feature library can quickly identify common defect types. The pre-set defect feature library contains the feature patterns of various known defects. By matching with the correlation matrix, it can be preliminarily judged whether there are defects in the components and the types of defects. The first detection result provides a basis for subsequent further analysis and processing. According to the first detection result, more detailed defect location, classification and evaluation can be carried out, providing a basis for formulating predictive maintenance strategies.

[0137] In this embodiment, the pre-set defect feature library stores the defect types of the equipment corresponding to the correlation matrix between the temperature distribution characteristics and the mechanical deformation characteristics, and the first detection result includes the component defect type.

[0138] Step 3: Modify the correlation matrix according to the first detection result and the component digital twin model and locate the abnormal area.

[0139] Step 3.1: Construct a digital twin model.

[0140] In this embodiment, the construction method of the component digital twin model includes:

[0141] Step 3.1.1: Based on the LiDAR point cloud data, use the hybrid modeling technology of B-Rep and CSG to construct a device digital twin model with LOD4-level accuracy.

[0142] LiDAR point cloud data can provide rich three-dimensional geometric information on the surface of the components. Based on these data, using the hybrid modeling technology of B-Rep (Boundary Representation) and CSG (Constructive Solid Geometry), a device digital twin model with LOD4 (Level of Detail 4, the highest level of detail) -level accuracy can be constructed. In the operation and maintenance management of substation components, the component digital twin model can be used for the coupling analysis of multiple physical fields to more deeply understand the performance and state of the components.

[0143] Step 3.1.2: Preset a fault model in the component digital twin model, and the fault model is used to simulate the behavior characteristics of the components under different component defect types.

[0144] The preset fault model can simulate the behavior characteristics of the components under different defect types in the digital twin model, such as overheating, deformation, insulation damage, etc. Through the simulation analysis of these fault models, the problems that may occur in the actual operation of the components can be predicted, and maintenance strategies can be formulated in advance to avoid the occurrence of faults.

[0145] Step 3.1.3: Use Kalman filtering to dynamically synchronize the behavioral characteristics of the components to the component digital twin model based on the monitoring data of the UAV.

[0146] Kalman filtering can perform real-time estimation and prediction of the behavioral characteristics of components based on the monitoring data of the UAV. Through Kalman filtering, the noise in the monitoring data can be filtered out to obtain more accurate device status information and dynamically synchronize it to the digital twin model. This dynamic update mechanism enables the model to capture minor changes in the component status, timely detect potential fault hazards, and provide more timely and accurate information for predictive maintenance. By synchronizing the behavioral characteristics of components in real time, the digital twin model can more accurately simulate the future development trend of components. Combining the fault model and prediction algorithm can improve the prediction accuracy of the occurrence time and severity of component faults, providing more powerful support for operation and maintenance decisions.

[0147] Step 3.2: Modify the parameters of the correlation matrix and locate the abnormal area.

[0148] In this embodiment, modifying the parameters of the correlation matrix and locating the abnormal area according to the first detection result and the component digital twin model includes:

[0149] Step 3.2.1: Map the first detection result to the fault model of the component digital twin model, simulate the behavioral characteristics of the component under the first detection result, and obtain the actual monitoring data of the component digital twin model.

[0150] Mapping the first detection result to the fault model of the component digital twin model can simulate the actual behavioral characteristics of the component under the detected potential defects. This helps to deeply understand the operating rules of the component in the fault state and provides a more intuitive and accurate basis for subsequent difference analysis and fault diagnosis.

[0151] By simulating to obtain the actual monitoring data of the component digital twin model, these data are generated considering the potential defects reflected by the first detection result. Comparing with the actual monitoring data, the status of the component can be more comprehensively evaluated, and possible problems can be found.

[0152] The simulation process can verify the rationality of the first detection result. If the simulated behavioral characteristics match the actual experience or theoretical expectations, it indicates that the first detection result has a certain reliability; otherwise, it may be necessary to re-examine the detection process and results.

[0153] Step 3.2.2: Calculate the deviation of the temperature distribution characteristics and the deviation of the mechanical deformation characteristics between the actual monitoring data and the expected monitoring data mapped by the first detection result to the component digital twin model to obtain the result of the difference analysis.

[0154] Calculating the deviation of temperature distribution characteristics and the deviation of mechanical deformation characteristics can quantify the degree of difference between the actual monitoring data and the expected monitoring data. These deviation values intuitively reflect the inconsistency between the actual state and the simulated state of the component, providing specific quantitative indicators for subsequent fault diagnosis and correction.

[0155] Considering the deviations of both temperature distribution characteristics and mechanical deformation characteristics simultaneously enables a comprehensive analysis of the component's state from multiple perspectives. Different defects may lead to different types of characteristic deviations, and through comprehensive analysis, the problems existing in the component can be judged more accurately.

[0156] The results of the difference analysis provide an important basis for operation and maintenance decisions. Based on the magnitude and type of the deviation, it can be determined whether further inspection, maintenance, or repair measures are required, and which measures are more appropriate to take.

[0157] Step 3.2.3: According to the results of the difference analysis, correct the parameters of the correlation matrix.

[0158] The correlation matrix is used to describe the relationship between temperature distribution characteristics and mechanical deformation characteristics. Correcting the parameters of the correlation matrix according to the results of the difference analysis can enable the correlation matrix to more accurately reflect the correlation relationship between the two characteristics under the actual state of the component, improving the accuracy and reliability of the model.

[0159] Different equipment defects and operating conditions may cause changes in the relationship between temperature distribution characteristics and mechanical deformation characteristics. By correcting the parameters of the correlation matrix, the defect detection model embedded with prior constraints of the component can be adapted to different operating conditions and defect types. A more accurate correlation relationship can improve the sensitivity and specificity of the defect detection model embedded with prior constraints of the component, reducing the situations of false detection and missed detection.

[0160] Step 3.2.4: Determine the abnormal area.

[0161] The abnormal area includes the temperature abnormal area and the deformation abnormal area.

[0162] If the deviation of the temperature distribution characteristics exceeds the preset temperature threshold, the area where the deviation of the temperature distribution characteristics occurs is taken as the temperature abnormal area.

[0163] If the deviation of the mechanical deformation characteristics exceeds the preset temperature threshold, the area where the deviation of the temperature distribution characteristics occurs is taken as the deformation abnormal area.

[0164] By setting temperature thresholds and deformation thresholds to determine temperature anomaly regions and deformation anomaly regions, it is possible to accurately locate the possible fault positions in the components, which helps the operation and maintenance personnel quickly find the problems and improve the maintenance efficiency. Determining the temperature anomaly regions and deformation anomaly regions separately can distinguish different types of abnormal situations. Different types of anomalies may correspond to different fault causes, which helps to more accurately diagnose the fault causes and take targeted maintenance measures.

[0165] Step 4: Input the corrected correlation matrix into the defect detection model embedded with prior constraints of the component, and output the second detection result.

[0166] In this embodiment, the prior constraints of the component include: temperature constraint of the component, geometric shape constraint of the component, position constraint of the component, material constraint of the component, and installation accuracy constraint of the component, which can screen and constrain the temperature distribution characteristics and mechanical deformation characteristics from multiple dimensions.

[0167] In this embodiment, the defect detection model embedded with prior constraints of the component includes:

[0168] The feature extraction module, including a temperature extraction branch and a deformation extraction branch, is used to extract the temperature distribution characteristics and mechanical deformation characteristics of the corrected correlation matrix and input them into the prior constraint embedding layer;

[0169] The prior constraint embedding layer, including the temperature constraint function of the component, the geometric shape constraint function of the component, the position constraint function of the component, the material constraint function of the component, and the installation accuracy function of the component, is used to screen the temperature distribution characteristics and mechanical deformation characteristics and input them into the dynamic sparse inference module;

[0170] The dynamic sparse inference module is used to select the temperature distribution characteristics and mechanical deformation characteristics whose relevance to defect detection exceeds the threshold and input them into the defect detection and classification module;

[0171] The defect detection and classification module is used for the YOLOv8 network to analyze according to the temperature distribution characteristics and mechanical deformation characteristics and output the second detection result;

[0172] In this embodiment, the second detection result includes the defect type, defect position, defect severity, and confidence level of the component.

[0173] Step 5: Construct a causal graph of the inspection record and the component working condition, and output a predictive maintenance strategy according to the second detection result and the anomaly region.

[0174] Step 5.1: Construct a causal graph of the inspection record and the component working condition.

[0175] Step 5.1.1: Establish an inspection record node and a component working condition node according to the inspection record and the component working condition.

[0176] Step 5.1.2: Based on the K2 algorithm with the maximum a posteriori probability as the optimization objective, perform heuristic search on the inspection nodes and component condition nodes according to domain knowledge to establish directed edges, and output a directed acyclic causal graph of inspection records and component conditions including the causal relationship strength weights.

[0177] In this embodiment, the directed edge is used to represent the causal relationship between the inspection node and the component condition node, the length of the directed edge is used to represent the causal strength between the inspection node and the component condition node, the inspection node includes the observed indicators in the inspection record, including temperature distribution and mechanical deformation, and the component condition node includes the state variables of the component condition, including normal operation, minor fault, moderate fault, and severe fault.

[0178] Step 5.2: Output a predictive maintenance strategy according to the second detection result and the abnormal area.

[0179] In this embodiment, the second detection result includes the defect type, defect location, defect severity, and confidence level of the component, and the abnormal area includes the shape and size of the abnormal part in the component.

[0180] Step 5.2.1: Assign risk weights to the defect type, defect location, defect severity, confidence level of the component, and the shape and size of the abnormal part in the component respectively, and calculate the risk score of the defect.

[0181] Assigning risk weights to the defect type, defect location, defect severity, confidence level of the component, and the shape and size of the abnormal part in the component can comprehensively and meticulously consider various factors affecting the operating state of the component. Different defect types may have different degrees of harm, the defect location may affect the key functions of the component, the defect severity directly reflects the damaged condition of the component, the confidence level reflects the reliability of the detection result, and the shape and size of the abnormal part are related to the spread and influence range of the fault. By reasonably assigning weights, the risk score of the defect can be accurately calculated to achieve a precise assessment of the component defect risk.

[0182] The calculation of the risk score quantifies various characteristics of the defect, enabling the operation and maintenance personnel to intuitively understand the risk degree of each defect, avoiding the uncertainty of subjective judgment, and providing a scientific basis for subsequent risk level classification and maintenance strategy formulation.

[0183] Step 5.2.2: According to the risk score of the defect, divide the risk level and formulate a predictive maintenance strategy.

[0184] Dividing the risk levels according to the risk scores of defects can clearly classify defects with different risk levels. For example, defects with higher risk scores can be classified as high-risk levels, those with moderate risk scores as medium-risk levels, and those with lower risk scores as low-risk levels, which helps the operation and maintenance personnel quickly identify and handle high-risk defects and reasonably arrange maintenance resources.

[0185] Different risk levels correspond to different maintenance priorities and handling methods. Through scientific risk level division, the substation can formulate predictive maintenance strategies according to its own operation and maintenance capabilities and actual needs, ensure that key components and important defects are processed in a timely manner, and at the same time avoid over-maintenance of low-risk defects and improve the operation and maintenance efficiency.

[0186] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code therein. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc. These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device realizes the functions specified in Figure 1 a process or multiple processes and / or boxes Figure 1 specified in one box or multiple boxes.

[0187] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A defect detection method for substation UAV inspection equipment, characterized in that, Including: Collecting the monitoring data of the drone; Performing temperature field inversion correction on the monitoring data, establishing a correlation matrix according to sub-pixel displacement quantization of mechanical deformation, and matching a preset defect feature library to obtain a first detection result; Correcting the correlation matrix according to the first detection result and the component digital twin model and locating the abnormal area; Inputting the corrected correlation matrix into a defect detection model embedded with prior constraints of the component to output a second detection result; Constructing a causal graph of the inspection record and the component working condition, and outputting a predictive maintenance strategy according to the second detection result and the abnormal area.

2. The defect detection method for the substation UAV inspection equipment according to claim 1, wherein, The monitoring data includes visible light images, infrared images and LiDAR point cloud data; among them, the method for collecting the monitoring data of the drone includes: Controlling a drone equipped with a multi-spectral sensor, an infrared thermal imager and a LiDAR sensor to inspect the substation according to a preset route; Synchronizing the acquisition clocks of the multi-spectral sensor, the infrared thermal imager and the LiDAR sensor by using the GPS PPS pulse signal; In an environment with electromagnetic interference, adopting an adaptive frequency hopping communication technology to switch the 2.4G / 5.8G / LoRa frequency bands for data transmission, and collecting visible light images, infrared images and LiDAR point cloud data of the component; Introducing an adaptive weight mechanism into the point-to-point distance error function to improve the iterative closest point ICP algorithm; Constructing a three-dimensional coordinate system of the component based on the LiDAR point cloud data, and using the improved iterative closest point ICP algorithm to map the pixels of the visible light image and the infrared image to a unified spatial coordinate system to obtain the monitoring data of the drone.

3. The defect detection method for the substation UAV inspection equipment according to claim 2, characterized in that, Constructing a three-dimensional coordinate system of the component based on the LiDAR point cloud data, and using the improved iterative closest point ICP algorithm to map the pixels of the visible light image and the infrared image to a unified spatial coordinate system to obtain the monitoring data of the drone, including: Based on the LiDAR point cloud data, calculating the geometric center of the component as the coordinate origin, using principal component analysis PCA to fit the surface normal vector of the component as the coordinate Z-axis, aligning the coordinate X-axis to the due north direction of the substation building direction, and establishing a three-dimensional coordinate system of the component; Using the improved iterative closest point ICP algorithm to calculate the transformation matrix between the visible light image and the infrared image in the three-dimensional coordinate system of the component; Adopting the SIFT algorithm to extract the scale-invariant feature points of the visible light image and the infrared image, and performing preliminary matching through a FLANN matcher to obtain matching point pairs; Generating a depth map by using the LiDAR point cloud data, and eliminating the mismatched points in the matching point pairs by verifying the geometric consistency of the component to obtain the corrected matching point pairs; Calculating the centroid coordinates of the corrected matching points, and performing singular value decomposition on the covariance matrix of the centroid coordinates to solve the optimal values of the rotation matrix and the translation vector; Optimizing the transformation matrix by using the optimal values of the rotation matrix and the translation vector, and mapping the pixels of the visible light image and the infrared image to a unified spatial coordinate system through the inverse projection formula to obtain the monitoring data of the drone.

4. The defect detection method for the substation UAV inspection equipment according to claim 3, characterized in that Performing temperature field inversion correction on the monitoring data, establishing a correlation matrix according to sub-pixel displacement quantization of mechanical deformation, and matching a preset defect feature library to obtain a first detection result, including: Measure the reflectivity and emissivity of the UAV material at different temperatures to pre-construct a reflectivity-emissivity curve; Based on the UAV material and the reflectivity-emissivity curve, obtain the reflectivity and emissivity data of the UAV material; According to the infrared image, use the Planck inverse function method to preliminarily invert the temperature field of the monitoring area, and convert the infrared radiation intensity into temperature distribution characteristics by using the obtained reflectivity and emissivity data of the UAV material to obtain the preliminary inversion result of the temperature field; Establish a correction model based on the influence of the environmental parameters of the monitoring area and the UAV flight attitude on the temperature field inversion, and correct the preliminary inversion result to obtain the temperature field inversion result; Register the visible light image and the infrared thermal image in the monitoring data of the UAV collected at different temperatures according to the temperature field inversion result, and calculate the sub-pixel displacement between the images; According to the sub-pixel displacement between the images and the UAV material, establish a mathematical model between the mechanical deformation and the sub-pixel displacement, and convert the sub-pixel displacement between the images into mechanical deformation characteristics; Construct a correlation matrix between the temperature distribution characteristics and the mechanical deformation characteristics, and match the correlation matrix with a preset defect feature library to obtain a first detection result; The preset defect feature library stores the defect types of the equipment corresponding to the correlation matrix between the temperature distribution characteristics and the mechanical deformation characteristics; The first detection result includes the component defect type.

5. The defect detection method for the substation UAV inspection equipment according to claim 4, characterized in that The method for constructing the component digital twin model includes: Based on the LiDAR point cloud data, use the B-Rep and CSG hybrid modeling technology to construct a digital twin model of the LOD4-level precision equipment; Preset a fault model in the component digital twin model, and the fault model is used to simulate the behavior characteristics of the component under different component defect types; Use the Kalman filter to dynamically synchronize the behavior characteristics of the component to the component digital twin model according to the monitoring data of the UAV.

6. The defect detection method for the substation UAV inspection equipment according to claim 5, characterized in that, According to the first detection result and the component digital twin model, correct the parameters of the correlation matrix and locate the abnormal area, including: Map the first detection result to the fault model of the component digital twin model, simulate the behavior characteristics of the component under the first detection result, and obtain the actual monitoring data of the component digital twin model; Calculate the temperature distribution characteristic deviation and the mechanical deformation characteristic deviation between the actual monitoring data and the expected monitoring data mapped by the first detection result to the component digital twin model to obtain the result of the difference analysis; According to the result of the difference analysis, correct the parameters of the correlation matrix; The abnormal area includes a temperature abnormal area and a deformation abnormal area; If the temperature distribution characteristic deviation exceeds the preset temperature threshold, the area where the temperature distribution characteristic deviation occurs is used as the temperature abnormal area; If the mechanical deformation characteristic deviation exceeds the preset temperature threshold, the area where the temperature distribution characteristic deviation occurs is used as the deformation abnormal area.

7. The defect detection method for the substation UAV inspection equipment according to claim 6, characterized in that, The component prior constraints include: the temperature constraint of the component, the geometric shape constraint of the component, the position constraint of the component, the material constraint of the component, and the installation accuracy constraint of the component.

8. The defect detection method for the substation UAV inspection equipment according to claim 7, characterized in that The defect detection model embedded with component prior constraints includes: The feature extraction module, including a temperature extraction branch and a deformation extraction branch, is used to extract the temperature distribution features and mechanical deformation features of the corrected correlation matrix and input them into the prior constraint embedding layer; The prior constraint embedding layer, including the temperature constraint function of the component, the geometric shape constraint function of the component, the position constraint function of the component, the material constraint function of the component, and the installation accuracy function of the component, is used to screen the temperature distribution features and mechanical deformation features and input them into the dynamic sparse inference module; The dynamic sparse inference module is used to select the temperature distribution features and mechanical deformation features whose relevance to defect detection exceeds the threshold and input them into the defect detection and classification module; The defect detection and classification module is used for the YOLOv8 network to analyze based on the temperature distribution features and mechanical deformation features and output the second detection result; The second detection result includes the defect type, defect location, defect severity, and confidence level of the component.

9. The defect detection method for the substation UAV inspection equipment according to claim 1, characterized in that, Construct a causal graph of the inspection record and the component working condition, including: Based on the inspection record and the component working condition, establish an inspection record node and a component working condition node; Based on the K2 algorithm with the maximum posterior probability as the optimization goal, conduct a heuristic search on the inspection node and the component working condition node according to domain knowledge to establish a directed edge, and output a directed acyclic causal graph of the inspection record and the component working condition containing the causal relationship strength weight; The directed edge is used to represent the causal relationship between the inspection node and the component working condition node; The length of the directed edge is used to represent the causal strength between the inspection node and the component working condition node; The inspection node includes the observation indicators in the inspection record, including temperature distribution and mechanical deformation amount; The component working condition node includes the state variables of the component working condition, including normal operation, minor fault, moderate fault, and severe fault.

10. The defect detection method for substation UAV inspection equipment according to claim 9, characterized in that, According to the second detection result and the abnormal area, output a predictive maintenance strategy, including: The second detection result includes the defect type, defect location, defect severity, and confidence level of the component; The abnormal area includes the shape and size of the abnormal part in the component; Allocate risk weights to the defect type, defect location, defect severity, confidence level of the component, and the shape and size of the abnormal part in the component respectively, calculate the risk score of the defect; according to the risk score of the defect, divide the risk level and formulate a predictive maintenance strategy.

Images