Power transmission line multi-component defect cooperative detection method and system suitable for unmanned aerial vehicle inspection image

Abstract

The invention provides a power transmission line multi-component defect cooperative detection method and system suitable for an unmanned aerial vehicle inspection image, and relates to the technical field of power transmission line intelligent inspection and image recognition, and the method comprises the steps: carrying out the three-dimensional geometric anchoring of a target component, and planning a scanning path; transmitting an energy beam to the surface of the target component, and collecting a physical response signal; converting the acquired physical response signal into a high-dimensional feature vector; constructing a dynamic response manifold of the target component; and calculating geodesic distances between points on the dynamic response manifold, and carrying out manifold geometric mutation analysis and defect positioning. According to the method, the defect type can be accurately identified, and defect diagnosis can be advanced from response after a fault to prediction before the fault.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection and image recognition technology for power transmission lines, specifically to a collaborative detection method and system for defects in multiple components of power transmission lines applicable to UAV inspection images. Background Technology

[0002] Power transmission lines are the lifeline of national energy transmission, and their safe and stable operation is of paramount importance. Key components on the lines, such as insulators, vibration dampers, hardware, and conductors, are exposed to complex natural environments for extended periods. These components are susceptible to various defects due to factors such as mechanical stress, electrical corrosion, material aging, and environmental erosion. In severe cases, these defects can lead to major accidents such as line flashovers and line breaks, causing enormous economic losses and social impacts. Therefore, regular, accurate, and efficient inspections of power transmission lines to promptly identify and eliminate potential hazards are the lifeline for ensuring the safety of the power grid.

[0003] Current power transmission line inspection technology has mainly gone through the following stages of development, but each stage has its own insurmountable limitations:

[0004] 1. Traditional manual inspection stage: This mainly relies on inspectors carrying tools such as binoculars to conduct ground patrols, or climbing poles after a power outage for close observation and inspection. The main drawbacks of this method are: poor safety, high labor intensity, and the need for inspectors to cope with complex terrain and work at heights, posing extremely high personal safety risks. It is also inefficient, with insufficient coverage and slow inspection speed, making it difficult to effectively cover poles in mountainous, forested, or river-crossing areas. Furthermore, it is highly subjective and has limited accuracy, with inspection results heavily dependent on the inspectors' experience and sense of responsibility, offering very limited ability to detect minor, hidden defects, and making quantitative assessment impossible.

[0005] 2. Passive Visual Inspection Stage Using Drones: With the development of drones and high-definition imaging technology, using drones equipped with visible light cameras, infrared thermal imagers, and ultraviolet imagers for inspection has become mainstream. This method captures high-definition images and videos, which are then analyzed manually or by artificial intelligence algorithms to identify defects. While this method significantly improves safety and efficiency, its core technological bottleneck lies in the superficial limitations of its detection capabilities. All inspections based on optical imaging, whether visible light, infrared, or ultraviolet, are essentially passive detections. They can only capture surface defects or secondary effects caused by defects. For numerous internal structural defects that do not show significant external changes or generate heat or ultraviolet radiation, such as early debonding between composite insulator core rods and sheaths, micro-cracks inside glass insulators, internal fatigue damage in metal fittings, and deterioration of the material's mechanical properties, this method is completely unable to detect them. Furthermore, it is highly dependent on the environment and lacks reliability. The accuracy of visual inspection is easily affected by environmental factors such as lighting conditions, shooting angle, shadows, and surface contamination, leading to high rates of missed and false alarms. The qualitative deficiencies in diagnostic information mean that even if surface cracks are found, their depth and impact on remaining mechanical strength cannot be quantified. Diagnostic results are mostly qualitative, indicating the presence or absence of defects, lacking quantitative assessment of the severity and development trend of defects, which is not conducive to refined condition-based maintenance decisions.

[0006] 3. Gaps and Challenges in Existing Technologies: In summary, existing technologies, whether traditional manual inspections or mainstream UAV visual inspections, are limited to passive observation of the surface morphology of components. The industry has consistently lacked an effective technical means to break through the surface and delve into the interior, to move from qualitative observation to quantitative diagnosis, and especially to detect early, hidden, and structural defects that pose a significant potential threat to line safety. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a collaborative detection method for defects in multiple components of power transmission lines based on UAV inspection images, solving the problem that existing technologies for detecting defects in power transmission lines are limited to appearance and cannot provide quantitative diagnosis.

[0008] Another objective of this invention is to provide a collaborative detection system for defects in multiple components of power transmission lines that is applicable to images inspected by unmanned aerial vehicles (UAVs).

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A collaborative detection method for defects in multiple components of power transmission lines using images from UAV inspections includes the following steps:

[0011] Construct a three-dimensional geometric model of the target component and plan a preset scanning path along the surface of the target component;

[0012] A non-contact energy excitation source emits an energy beam along a preset scanning path toward the surface of the target component and collects the physical response signal;

[0013] The acquired physical response signals are converted into high-dimensional feature vectors;

[0014] Using the three-dimensional geometric model of the target component as a basis, the high-dimensional feature vector is used as an additional dimension to construct the dynamic response manifold of the target component;

[0015] Calculate the geodesic distances between points on the dynamic response manifold, and perform manifold geometric abrupt change analysis and defect location based on the geodesic distances.

[0016] Preferably, in the step of constructing the three-dimensional geometric model of the target component, the three-dimensional geometric model is anchored 1:1 to the target component entity, the dense point cloud output by the lidar provides the geometric structure framework, the texture information captured by the visual camera is used for loop closure detection and elimination of long-term drift, and the high-frequency motion data provided by the inertial measurement unit (IMU) is coupled into the entire state estimator through an extended Kalman filter.

[0017] Preferably, in the step of transmitting energy beams, energy beams with a preset spectral width are transmitted at continuous spatial points along the path; in the step of acquiring physical response signals, physical response signals carrying time-domain information are acquired from each spatial excitation point.

[0018] Preferably, in the physical response signal conversion step, a short-time Fourier transform is performed on the original physical response signal of each spatial acquisition point. During the transformation, a Hanning window is selected as the window function to suppress spectral leakage. Features that can comprehensively characterize the physical state of the point are extracted from the generated graph and the original physical response signal and combined into a high-dimensional feature vector. The feature vector contains at least one set of feature parameters extracted from the frequency domain, time-frequency domain or modal space of the response signal.

[0019] Preferably, in the step of constructing the dynamic response manifold of the target component, the high-dimensional feature vector corresponding to each spatial point is used as the additional dimension of that point and embedded into the three-dimensional geometric model to form a two-dimensional or three-dimensional manifold in high-dimensional space that describes the continuous change of the physical response characteristics of the target component.

[0020] Preferably, in the geodesic distance calculation step, in the high-dimensional space composed of high-dimensional feature vectors, for each data point, the K-nearest neighbor algorithm is used to find its N nearest neighbor points, and the K value is set to N; the edge weight is set to the Euclidean distance between the neighbor points, and then the Dijkstra algorithm is applied to calculate the shortest path distance between all node pairs. This distance is an effective approximation of the geodesic distance between the two points.

[0021] Preferably, the manifold geometric mutation analysis and defect location step identifies regions on the manifold where geodesic distances are discontinuous as manifold anomaly regions. These anomaly regions correspond to internal or external defects on the component that alter the physical constitutive relationship. The manifold anomaly regions are then projected in reverse onto the three-dimensional geometric model of the target component to achieve defect location and visualization.

[0022] A collaborative defect detection system for multiple components of power transmission lines based on UAV inspection images includes:

[0023] The geometric model building module is used to build a three-dimensional geometric model of the target part and plan a preset scanning path along the surface of the target part.

[0024] The active flaw detection module includes a non-contact energy excitation source and a non-contact response sensor. The non-contact energy excitation source emits an energy beam toward the surface of the target component along a preset scanning path; the non-contact response sensor collects the physical response signal.

[0025] The feature extraction module converts the collected physical response signals into high-dimensional feature vectors;

[0026] The manifold construction module uses the three-dimensional geometric model of the target component as a basis and adds the high-dimensional feature vector as an additional dimension to construct the dynamic response manifold of the target component.

[0027] The defect diagnosis module calculates the geodesic distance between points on the dynamic response manifold and performs manifold geometric abrupt change analysis and defect location based on the geodesic distance.

[0028] Preferably, in the geometric model construction module, the three-dimensional geometric model is anchored 1:1 to the target component entity, the dense point cloud output by the lidar provides the geometric structure framework, the texture information captured by the visual camera is used for loop closure detection and elimination of long-term drift, and the high-frequency motion data provided by the inertial measurement unit (IMU) is coupled into the entire state estimator through an extended Kalman filter.

[0029] Preferably, in the active flaw detection module, the energy excitation source includes an acoustic excitation source and an electromagnetic excitation source. The acoustic excitation source emits a frequency-modulated ultrasonic beam, and the electromagnetic excitation source emits a swept-frequency electromagnetic pulse in the millimeter wave or terahertz frequency band. The response sensor includes a laser Doppler vibrometer and a coherent electromagnetic echo receiver. The laser Doppler vibrometer is used for non-contact measurement of the vibration response of the component surface caused by acoustic excitation. The coherent electromagnetic echo receiver is used to collect the absorption, reflection, or scattering response of the component material to electromagnetic waves of different frequencies.

[0030] Preferably, the feature extraction module performs a short-time Fourier transform on the original physical response signal of each spatial acquisition point. During the transform, a Hanning window is selected as the window function to suppress spectral leakage. Features that can comprehensively characterize the physical state of the point are extracted from the generated graph and the original physical response signal and combined into a high-dimensional feature vector. The feature vector contains at least one set of feature parameters extracted from the frequency domain, time-frequency domain or modal space of the response signal.

[0031] Preferably, the manifold construction module uses the high-dimensional feature vector corresponding to each spatial point as an additional dimension of that point, embeds it into a three-dimensional geometric model, and forms a two-dimensional or three-dimensional manifold in a high-dimensional space that describes the continuous change of the physical response characteristics of the target component.

[0032] Preferably, the defect diagnosis module calculates the geodesic distance. In a high-dimensional space composed of high-dimensional feature vectors, for each data point, the K-nearest neighbor algorithm is used to find its N nearest neighbor points, with K set to N. The edge weights are set to the Euclidean distance between neighbor points, and then the Dijkstra algorithm is applied to calculate the shortest path distance between all node pairs. This distance is an effective approximation of the geodesic distance between two points. Regions on the manifold where the geodesic distance is discontinuous are identified as manifold anomaly areas. The anomaly areas correspond to internal or external defects on the component that have altered physical constitutive relations. The manifold anomaly areas are then projected backward onto the three-dimensional geometric model of the target component to achieve defect localization and visualization.

[0033] The present invention has the following beneficial effects:

[0034] 1. This invention, through a novel paradigm of active energy excitation and response analysis, introduces the detection signal into the component. The feedback signal directly carries constitutive physical information such as the medium, density, and elastic modulus of the material. This invention can accurately identify defect types that are completely imperceptible by traditional methods, such as early poor bonding between the core rod and sheath inside composite insulators, microcrack propagation under the tempered layer of glass insulators, and internal fatigue damage to metal fittings caused by long-term vibration. It can detect early performance degradation of materials before macroscopic damage occurs, such as material embrittlement and decreased elastic modulus. This advances defect diagnosis from post-fault response to pre-fault prediction, providing key technical support for preventive maintenance.

[0035] 2. The core data collected by this invention is the inherent physical response of a component to a specific energy excitation. This response is a direct reflection of its internal physical state and has high stability and uniqueness. Regardless of whether it is under strong light, backlight, shadow, or when the component surface is covered with a thin layer of dust, the internal physical response characteristics remain basically unchanged, ensuring high consistency and repeatability of the detection results. Furthermore, for non-structural anomalies such as bird droppings and stains that are only attached to the surface, since they do not change the dynamic response characteristics of the component itself, they will not produce significant geometric abrupt changes in the manifold analysis of this invention, and can therefore be effectively filtered out, completely solving the persistent problem that visual algorithms easily misreport such situations as defects.

[0036] 3. This invention achieves quantitative analysis of defects by constructing a mathematical model of a dynamic response manifold. The physical severity of a defect is directly reflected in the degree of distortion of the manifold geometry. By calculating these geometric parameters, the severity of the defect can be transformed into a specific and quantifiable indicator. By periodically inspecting the same component, the changes in its dynamic response manifold over time can be tracked. By analyzing the evolution trend of the manifold geometric characteristics, the development speed of defects and the remaining lifespan of components can be predicted. This provides unprecedented data support for formulating optimal maintenance and replacement strategies and promotes the intelligent upgrade of power grid operation and maintenance from periodic maintenance to condition-based maintenance.

[0037] 4. In addition to diagnosing individual components, the technical framework of this invention can also diagnose systemic, non-physical damage-related collaborative state defects by analyzing the response characteristics of adjacent and related components. These defects include minor loosening at the connection between fittings and insulators, and attenuation of the gripping force of anti-vibration hammer clamps. These defects also pose a threat to line safety, but cannot be identified in traditional visual inspection because there are no obvious morphological changes. Attached Figure Description

[0038] Figure 1 This is a system architecture diagram of the present invention;

[0039] Figure 2 This is a cloud diagram illustrating the system composition of the present invention;

[0040] Figure 3 This is a flowchart of the process of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1

[0043] like Figures 1 to 3 As shown, a collaborative detection method for defects in multiple components of power transmission lines based on UAV inspection images includes the following steps:

[0044] The drone approaches the power transmission line component under test, quickly scans and reconstructs its three-dimensional surface geometry model using a vision or lidar system, and plans a preset scanning path along the surface of the target component.

[0045] The UAV travels along the scanning path and emits energy beams with a preset spectral width to continuous spatial points on the surface of the target component via an onboard non-contact energy excitation source. Simultaneously, an onboard non-contact response sensor synchronously collects the physical response signal carrying time-domain information fed back from each spatial excitation point.

[0046] The time-domain response signal at each spatial point is processed by signal transformation to convert it into a high-dimensional feature vector that can characterize the dynamic constitutive properties of that point; the feature vector contains at least one set of feature parameters extracted from the frequency domain, time-frequency domain or modal space of the response signal.

[0047] Using the three-dimensional geometric model of the component as a basis, the high-dimensional feature vectors obtained in the above steps, which correspond one-to-one with each spatial point, are used as additional dimensions of that point and embedded into the three-dimensional geometric model to construct a two-dimensional or three-dimensional manifold in a high-dimensional space that describes the continuous change of the physical response characteristics of the component, which is called a dynamic response manifold.

[0048] On the dynamic response manifold, the geometric smoothness of the manifold is analyzed by calculating the geodesic distance and local curvature between points on the manifold; regions on the manifold where the geodesic distance is discontinuous or the local curvature changes abruptly are identified as manifold anomaly regions, which correspond to internal or external defects on the component that have altered physical constitutive relations; finally, the manifold anomaly regions are projected in reverse onto the three-dimensional geometric model of the component to achieve precise location and visualization of defects.

[0049] Manifestations of defects:

[0050] Internal microcracks: These will form a sharp ridge on the manifold, which is a region where the curvature changes drastically.

[0051] Internal degumming / delamination: This causes changes in energy conduction paths. On the manifold, this manifests as two points that should be very close having an unusually large shortest distance within the manifold, as if a cliff has appeared in between, i.e., discontinuity in geodesic distance. Specific Implementation Example 2

[0053] like Figures 1 to 3As shown, the following is an architecture and core model of a collaborative detection system for multi-component defects in power transmission lines, applicable to UAV inspection images.

[0054] The complete implementation of this system relies on a closed-loop architecture that tightly couples a space-based detection system and a ground-based analysis system.

[0055] The space-based detection system is an airborne intelligent payload integrated into a high-performance flight platform.

[0056] Flight platform: A highly stable industrial-grade hexacopter UAV platform is selected. This platform is required to have a thrust-to-weight ratio of more than 2.5 to ensure sufficient maneuverability when carrying heavy payloads. In order to meet the requirements of long-term fine scanning operations, its effective operating endurance must be greater than 25 minutes after being equipped with a complete set of detection systems. The platform's navigation system integrates real-time dynamic differential positioning technology, i.e., RTK technology, to ensure that its three-dimensional spatial positioning accuracy is better than 2 centimeters horizontally and 3 centimeters vertically.

[0057] Geometric Model Construction Module: This module serves as the spatial reference for all subsequent detection data. It consists of a solid-state LiDAR and a global shutter high-resolution camera. The LiDAR has 128 laser lines and can generate more than 200,000 point cloud data per second, achieving a 360-degree horizontal scan without blind spots. Working in conjunction with it, the high-resolution camera sensor has more than 24 million pixels and uses a global shutter to eliminate image ghosting during the UAV's movement, ensuring that clear and sharp texture information is obtained during flight.

[0058] Active flaw detection module: This is the core of the system and is responsible for performing physical detection tasks.

[0059] Energy excitation source: A phased array ultrasonic transducer array consisting of 128 piezoelectric elements is used, with its center operating frequency set at 1.5 MHz. By precisely controlling the phase of the signal emitted by each array element, a focused sound beam that can be electronically deflected within a range of ±45 degrees can be generated.

[0060] Response sensor: A high-sensitivity laser Doppler vibration meter is used. The effective working distance of the instrument is one to three meters, and the velocity measurement range can reach ±10 meters per second. The signal acquisition bandwidth of its internal photodetector is greater than five megahertz, which is sufficient to capture all high-frequency vibration information caused by ultrasonic excitation.

[0061] The feature extraction module converts the collected physical response signals into high-dimensional feature vectors.

[0062] The manifold construction module uses the three-dimensional geometric model of the target component as a basis and adds the high-dimensional feature vector as an additional dimension to construct the dynamic response manifold of the target component.

[0063] The defect diagnosis module calculates the geodesic distance between points on the dynamic response manifold and performs manifold geometric mutation analysis and defect location.

[0064] Payload control and synchronization: This is the brain of the airborne system. Its core processor is NVIDIA's Jetson AGX Orin high-performance computing module, which is responsible for running complex real-time mapping and sensor control algorithms. Connected to this module is a field-programmable gate array, or FPGA, which is responsible for providing nanosecond-level precision synchronization hardware trigger signals for the ultrasonic emission and laser vibration meter acquisition actions through hardware logic circuits.

[0065] The ground-based analysis system is a suite of software deployed on a high-performance graphics workstation, responsible for data analysis and decision-making.

[0066] Task planning software module: Provides a graphical interface that allows operators to plan tasks on a 3D geographic information system and select and load standard dynamic response manifold baseline models of target models from the component database.

[0067] Data processing and analysis engine: This is the core software that integrates a series of algorithms, including signal processing, machine learning, and manifold geometry calculation.

[0068] Visualization and Reporting System: Responsible for transforming abstract manifold analysis results into 3D models and structured reports that engineers can intuitively understand.

[0069] Detailed Explanation of Core Models and Algorithms:

[0070] 3D Geometric Anchoring Model: This scheme adopts real-time vision-inertial-liDAR simultaneous localization and mapping technology based on multi-sensor fusion. The dense point cloud output by the LiDAR provides a robust geometric framework, while the rich texture information captured by the vision camera is used for efficient loop closure detection and elimination of long-term drift. The high-frequency motion data provided by the inertial measurement unit (IMU) is tightly coupled into the entire state estimator through an extended Kalman filter, which greatly improves the system's response capability to rapid maneuvers. The final output of this model is a triangular mesh model of the component surface with real physical scale, stored in PLY format.

[0071] Excitation Signal and Control: The excitation signal uses a linear frequency modulated pulse, commonly known as a chirp signal. The goal of selecting this signal mode is to inject wideband energy within a limited transmission time, thereby achieving high signal-to-noise ratio and high time resolution at the receiver through pulse compression technology. In this scheme, the chirp signal has a pulse duration T of 100 microseconds, a sweep bandwidth B of 1 MHz, and the signal frequency increases linearly from 1.0 MHz to 2.0 MHz. The phased array beam control algorithm dynamically calculates the excitation phase delay of each array element based on the distance and angle information returned by the lidar in real time, ensuring that the beam focus always falls precisely on the current target point on the scanning path.

[0072] Dynamic Response Feature Extraction: This model is responsible for transforming the raw time-domain vibration signals acquired by the laser Doppler vibrometer into meaningful physical features. First, a short-time Fourier transform is performed on the raw signal at each spatial acquisition point. During the transform, a Hanning window is used as the window function to effectively suppress spectral leakage. The window function length is set to 256 sampling points, and the overlap rate between adjacent windows is set to 75%. This set of parameters represents the experimental result that achieves the best balance between time and frequency resolution. Next, features that comprehensively characterize the physical state of a point are extracted from the generated graph and the raw signal through a series of algorithms and combined into a 50-dimensional feature vector. The feature vector contains at least one set of feature parameters extracted from the frequency domain, time-frequency domain, or modal space of the response signal. Modal Frequency: The top five strongest resonant frequencies are identified by applying a peak-finding algorithm to the power spectrum after time averaging. Damping Ratio: After narrowband filtering of the signal at each modal frequency, the attenuation rate of its amplitude over time is calculated using the logarithmic decay method. Power Spectral Density: The power spectrum is integrated within specific high-frequency and low-frequency bands to obtain the total energy value for each band. Guided wave group velocity: The composite insulator sheath can be regarded as a waveguide structure. Ultrasonic excitation will generate guided waves such as Lamb waves in it. By performing cross-correlation calculation on the received signals of two spatially adjacent detection points, the flight time difference of a specific wave packet can be accurately measured, thereby calculating the group velocity of the guided wave. Defects such as internal debonding will cause wave reflection and mode conversion, resulting in significant changes in the group velocity.

[0073] Dynamic response manifold construction and analysis: This scheme uses the isometric mapping, i.e., the Isomap algorithm, to construct and analyze manifolds.

[0074] Adjacency graph construction: In a high-dimensional space consisting of fifty feature vectors, for each data point, the K-nearest neighbor algorithm is used to find its ten nearest neighbor points. The value of K is set to ten, which is an optimal choice that balances capturing the local geometric details of the manifold and suppressing the influence of sensor noise.

[0075] Geodesic distance calculation: On the adjacency graph constructed in the previous step, set the edge weight to the Euclidean distance between neighboring points, and then apply Dijkstra's algorithm to calculate the shortest path distance between all pairs of nodes in the graph. This distance is an effective approximation of the geodesic distance between two points.

[0076] Analysis and Visualization: Finally, using classic multidimensional scaling techniques, the matrix containing all point-to-geodesic distances is projected into a three-dimensional space for visualization. The geometry of this three-dimensional point cloud then intuitively presents the dynamic response manifold. Specific Implementation Example 3

[0078] like Figures 1 to 3 As shown, the following is a judgment scheme and data processing method for a collaborative detection system of multi-component defects in power transmission lines, applicable to UAV inspection images:

[0079] System decision-making scheme and control logic:

[0080] System control logic: The payload control unit, acting as a slave, sends a heartbeat packet containing sensor status and target distance to the flight control master at a frequency of 10 Hz. The flight control master then performs real-time position fine-tuning based on this packet. When the lidar ranging value is continuously less than the safety threshold of 0.8 meters, or when the payload control unit reports an abnormal sensor data stream, the system will trigger a safety hovering and alarm mechanism, awaiting manual takeover.

[0081] System determination scheme:

[0082] Level 1 Judgment (Anomaly Detection):

[0083] Metric: Local Geodesic Distortion Rate (LGD), which is precisely defined as the difference between the sum of squared geodesic distances between a point and its neighbors and the sum of squared Euclidean distances in the low-dimensional embedding space.

[0084] Logic: The threshold T1 for judging LGD score is set as the 99th percentile of the LGD score of all points on the healthy standard manifold. Any point with a score exceeding T1 is marked as abnormal.

[0085] Secondary judgment (defect classification):

[0086] Model: A support vector machine classifier with a radial basis function kernel is used. This kernel function is chosen because it performs well in processing nonlinearly separable data composed of manifold geometric features.

[0087] Logic: Input the geometric features of the abnormal region, the principal axis length ratio after principal component analysis, the region area, the maximum value of LGD score, etc. into the classifier, and output the defect category.

[0088] Level 3 Judgment (Severity Level Assessment):

[0089] Indicator: Comprehensive Defect Index (CDI), which is calculated as CDI = 0.4 * log(Area) + 0.5 * max(LGD_Score) + 0.1 * Curvature_Estimate, where the weighting coefficients are derived from regression calibration of a large amount of laboratory destructive test data.

[0090] Logical approach: Based on the CDI value, a four-level classification is adopted. A CDI value between 0 and 2.5 is classified as Level I concern, and a value greater than 7.0 is classified as Level IV critical.

[0091] Data content processing and input / output:

[0092]

[0093]

[0094] Application Examples

[0095] Scenario: A ±500 kV ultra-high voltage direct current transmission line in a region in Northwest China. The line passes through the Gobi Desert, where it is subject to huge temperature differences between day and night and wind and sand erosion all year round, which puts extremely high demands on the mechanical and thermal stress of the insulation components.

[0096] Challenge: Traditional drone visible light and infrared inspections have failed to detect any abnormalities after multiple inspections. However, based on historical data analysis, the maintenance department found that the glass insulators in this section had a slightly higher rate of spontaneous explosion for unknown reasons, and suspected that there was early internal damage that could not be detected by the outside.

[0097] On-site operation and execution process:

[0098] Task Planning: The maintenance team selected thirty towers in this section as the targets for this detailed inspection. The standard dynamic response manifold reference model for the LXP-160 glass insulators used on this line was loaded into the ground-based analysis system.

[0099] Data Acquisition: The space-based detection system autonomously flies to the suspension point of the insulator string on the target tower. At a distance of 1.5 meters from the surface of the insulator string, it activates the three-dimensional geometric anchoring unit and completes the construction of the three-dimensional mesh model of the entire insulator string in about 30 seconds. Then, the UAV descends along the spiral path of the insulator string at a speed of 0.1 meters per second. The active flaw detection unit performs high-density scanning on the surface of the shed of each glass insulator. The complete scanning and data acquisition of a single insulator string takes about four minutes.

[0100] Data analysis and system judgment:

[0101] Manifold Construction: After receiving the feedback data from a string of insulators in phase B of tower N47, the ground-side analysis engine automatically completed feature extraction and dynamic response manifold construction.

[0102] Anomaly detected: When analyzing the manifold corresponding to the twelfth insulator, the system found that it presents a very smooth geometric surface overall, but in a region near the cement bonded part of the steel cap, the geometry of the manifold shows an extremely sharp, knife-like linear ridge.

[0103] Quantitative Diagnosis: System Judgment Scheme Activation:

[0104] Level 1 Judgment: The local geodesic distortion rate (LGD) score of this ridge area soared to 9.8, far exceeding the anomaly threshold T1 set at 3.5, and was therefore judged as an anomaly area.

[0105] Secondary judgment: The geometric shape of the abnormal area is a highly linear, narrow band with a highly consistent curvature gradient direction. Based on this feature, the support vector machine classifier diagnoses it as an internal structural crack with a confidence level of 98%.

[0106] Level 3 Judgment: The Comprehensive Defect Index (CDI) is calculated to be as high as 8.2, exceeding the critical threshold of 7.0. The system finally outputs the diagnostic conclusion: There is a critical level (Level IV) microcrack inside the No. 12 insulator, which has an extremely high risk of spontaneous explosion.

[0107] Verification and final results:

[0108] Verification measures: Based on the report, the maintenance department immediately arranged live-line work to replace the insulator. The replaced insulator was sent to the Electric Power Research Institute for laboratory testing. Through high-precision industrial CT scanning, a seven-millimeter-long internal radial crack extending from the base of the steel cap into the interior of the glass component was found in a location completely consistent with the system report.

[0109] Beneficial effects: This test successfully predicted a highly likely UHV line insulator self-explosion accident, avoiding a serious fault that would have caused a single-pole blockage in the DC system, and saving huge transmission losses and the risk of grid instability. This case fully demonstrates the penetrating diagnostic capability of this technology for detecting invisible, high-risk internal structural defects.

[0110] This invention also provides a storage medium storing a computer program. When executed by a processor, the computer program implements some or all of the steps in various embodiments of the collaborative detection method for multi-component defects in power transmission lines based on UAV inspection images provided by this invention. The storage medium may be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0111] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0112] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by an inclusion clause does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0113] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for cooperative detection of multi-component defects of transmission lines suitable for unmanned aerial vehicle (UAV) inspection images, characterized in that, The method comprises the following steps: constructing a three-dimensional geometric model of the target component, and planning a preset scanning path along the surface of the target component; a non-contact energy excitation source emits an energy beam to the surface of the target component along the preset scanning path, and collects a physical response signal; the collected physical response signal is converted into a high-dimensional feature vector; the three-dimensional geometric model of the target component is taken as a base, and the high-dimensional feature vector is taken as an additional dimension to construct a dynamic response manifold of the target component; geodesic line distances between each point on the dynamic response manifold are calculated, and manifold geometric mutation analysis and defect positioning are performed according to the geodesic line distances. 2.The method of claim 1, wherein In the step of constructing the three-dimensional geometric model of the target component, the three-dimensional geometric model is anchored with the target component entity in a 1:1 manner, dense point clouds output by a laser radar provide a geometric structure framework, texture information captured by a visual camera is used for loop detection and long-term drift elimination, and high-frequency motion data provided by an inertial measurement unit (IMU) is coupled into the entire state estimator through an extended Kalman filter. 3.The method of claim 1, wherein In the step of emitting the energy beam, an energy beam with a preset spectral width is emitted to continuous spatial points on the path; and in the step of collecting the physical response signal, a physical response signal carrying time domain information is collected from each spatial excitation point.

4. The method according to claim 1, characterized in that, In the step of converting the physical response signal, short-time Fourier transform is performed on the original physical response signal of each spatial collection point, and a Hanning window is selected as a window function to suppress spectral leakage during the transform process; features capable of comprehensively representing the physical state of the point are extracted from the generated graph and the original physical response signal, and are combined into a high-dimensional feature vector; the feature vector contains at least one group of feature parameters extracted from the frequency domain, time-frequency domain or modal space of the response signal.

5. The method according to claim 1, wherein, In the step of constructing the dynamic response manifold of the target component, the high-dimensional feature vector corresponding to each spatial point is taken as an additional dimension of the point, is embedded into the three-dimensional geometric model, and a two-dimensional or three-dimensional manifold describing the continuous change of the physical response characteristics of the target component in a high-dimensional space is formed.

6. The method according to claim 1, wherein, In the step of calculating the geodesic line distance, in the high-dimensional space composed of the high-dimensional feature vectors, for each data point, K-nearest neighbor algorithm is used to find its N nearest neighbor points, and the value of K is set as N; the edge weight is set as the Euclidean distance between the neighbor points, and then Dijkstra algorithm is applied to calculate the shortest path distance between all node pairs, which is an effective approximation of the geodesic line distance between the two points.

7. The method according to claim 1, wherein, In the step of manifold geometric mutation analysis and defect positioning, an area where the geodesic line distance is discontinuous on the manifold is identified as an abnormal area of the manifold, and the abnormal area corresponds to an internal or external defect existing on the component where the physical constitutive relation is changed; the abnormal area of the manifold is reversely projected onto the three-dimensional geometric model of the target component to realize the positioning and visualization of the defect.

8. The detection system for the power line multi-component defect collaborative detection method suitable for the unmanned aerial vehicle inspection image according to any one of claims 1-7, characterized in that, The method comprises the following steps: a geometric model construction module is configured to construct a three-dimensional geometric model of a target component, and plan a preset scanning path along the surface of the target component; an active flaw detection module comprises a non-contact energy excitation source and a non-contact response sensor, and the non-contact energy excitation source emits an energy beam to the surface of the target component along the preset scanning path; The non-contact response sensor collects a physical response signal; The feature extraction module converts the collected physical response signal into a high-dimensional feature vector; The manifold construction module constructs a dynamic response manifold of the target component by taking the three-dimensional geometric model of the target component as a base and taking the high-dimensional feature vector as an additional dimension. The defect diagnosis module calculates the geodesic distance between each point on the dynamic response manifold, and performs manifold geometric mutation analysis and defect positioning according to the geodesic distance.

9. The power line multi-component defect cooperative detection system for unmanned aerial vehicle inspection image according to claim 8, characterized in that, The geometric model construction module is anchored with the target component entity in a 1:1 manner, uses the dense point cloud output by the laser radar to provide a geometric structure framework, uses the texture information captured by the visual camera for loop detection and long-term drift elimination, and uses the high-frequency motion data provided by the inertial measurement unit (IMU) to be coupled into the entire state estimator through an extended Kalman filter.

10. The power line multi-component defect cooperative detection system for unmanned aerial vehicle inspection image according to claim 8, characterized in that, In the active detection module, the energy excitation source includes an acoustic excitation source and an electromagnetic excitation source, the acoustic excitation source emits a frequency-modulated ultrasonic beam, and the electromagnetic excitation source emits a swept-frequency electromagnetic pulse in the millimeter wave or terahertz band; the response sensor includes a laser Doppler vibrometer and a coherent electromagnetic echo receiver, the laser Doppler vibrometer is used for non-contact measurement of the vibration response of the component surface caused by acoustic excitation, and the coherent electromagnetic echo receiver is used for collecting the absorption, reflection or scattering response of the component material to electromagnetic waves of different frequencies.

11. The power line multi-component defect cooperative detection system for unmanned aerial vehicle inspection image according to claim 8, characterized in that, The feature extraction module performs short-time Fourier transform on the original physical response signal of each spatial collection point, and selects a Hanning window as a window function to suppress spectral leakage during the transform process; from the generated semantic graph and the original physical response signal, features that can comprehensively represent the physical state of the point are extracted to form a high-dimensional feature vector; the feature vector contains at least one group of feature parameters extracted from the frequency domain, time-frequency domain or modal space of the response signal.

12. The power line multi-component defect cooperative detection system for unmanned aerial vehicle inspection image according to claim 8, characterized in that, The manifold construction module embeds the high-dimensional feature vector corresponding to each spatial point as an additional dimension of the point into the three-dimensional geometric model to form a two-dimensional or three-dimensional manifold in a high-dimensional space that describes the continuous change of the physical response characteristics of the target component.

13. The power line multi-component defect cooperative detection system for unmanned aerial vehicle inspection image according to claim 8, characterized in that, The defect diagnosis module calculates the geodesic distance, in a high-dimensional space composed of high-dimensional feature vectors, for each data point, uses the K-nearest neighbor algorithm to find its N nearest neighbor points, and sets the K value to N; sets the edge weight as the Euclidean distance between the neighbor points, and then applies the Dijkstra algorithm to calculate the shortest path distance between all node pairs, which is an effective approximation of the geodesic distance between the two points; an area where the geodesic distance is discontinuous on the manifold is identified as an abnormal area of the manifold, and the abnormal area corresponds to an internal or external defect that changes the physical constitutive relation of the component; the abnormal area of the manifold is reversely projected onto the three-dimensional geometric model of the target component to realize the positioning and visualization of the defect.

14. A computer device, comprising: It comprises: a processor; and a memory configured to store machine-readable instructions that, when executed by the processor, perform the method for collaborative detection of multi-component defects of a power transmission line image suitable for unmanned aerial vehicle inspection as claimed in any one of claims 1-7.

15. A storage medium, characterized by The storage medium stores a computer program, and the computer program is executed by the processor to implement the power transmission line multi-component defect collaborative detection method suitable for unmanned aerial vehicle image inspection according to any one of claims 1-7.

Images