Autonomous Detection and Repair Methods and Systems for Damaged Insulated Overhead Lines

By combining infrared, vision, and vibration sensors with multimodal data processing via convolutional neural networks, high-precision autonomous detection and repair of insulated overhead lines have been achieved, solving the problems of insufficient detection accuracy and unstable repair quality, and realizing full automation and quantitative repair.

CN122860027BActive Publication Date: 2026-11-10常州金坛金能电力有限公司
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611363256.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-10
Estimated Expiration
2046-09-04

AI Technical Summary

Technical Problem

In existing technologies, the detection accuracy of insulated overhead lines is insufficient, the repair quality is unstable, and the degree of automation is low, making it difficult to achieve efficient autonomous detection and repair.

Method used

Infrared sensors, vision sensors, and vibration sensors are used to collect multimodal data in real time. A convolutional neural network model is used to calculate the comprehensive confidence level of damage. Combined with 3D measurement, autonomous detection and repair are achieved, including the use of unified data mapping, polarization filter preprocessing, deformable convolutional layers, and coordinate attention modules.

Benefits of technology

It achieves high-precision damage detection and stable repair quality, and can be fully automated to obtain quantitative dimensional information, avoiding the high-risk operation of traditional manual inspection and the limitation of single detection of existing inspection robots.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122860027B_ABST
    Figure CN122860027B_ABST
Patent Text Reader

Abstract

This invention relates to the field of power equipment maintenance technology, and particularly to a method and system for autonomous detection and repair of damage to insulated overhead lines. The method includes the following steps: timestamping the data collected by infrared sensors, vision sensors, and vibration sensors, and mapping the infrared pixel coordinates, image pixel coordinates, and partial discharge source coordinates to the robot's base coordinate system; calculating the temperature increment within the sampling time window; inputting the visual image within the time window into a convolutional neural network model, outputting the crack confidence score and mask region; performing frequency domain analysis on the vibration signal within the time window to calculate the vibration anomaly score; calculating the overall damage confidence score based on the temperature increment, crack confidence score, and vibration anomaly score; if the overall damage confidence score is not less than the damage threshold, it is determined to be a real damage, triggering 3D measurement and performing repair. This invention integrates "detection-judgment-repair-verification" into one system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power equipment maintenance technology, and in particular to a method and system for autonomous detection and repair of damage to insulated overhead lines. Background Technology

[0002] Overhead power distribution lines are prone to insulation damage due to aging, lightning strikes, and tree short circuits caused by prolonged exposure to the outdoors, thus requiring regular maintenance. Traditional maintenance methods rely mainly on manual visual inspection or single image recognition, which is insufficient to detect minute cracks and internal partial discharges, lacking quantitative dimensional information. Furthermore, manual pole climbing for insulation wrapping results in inconsistent tightness and overlap, low standardization, unstable repair quality, and is a high-risk, high-altitude operation.

[0003] Currently, there are inspection robots on the market for damage detection, but they can only detect, not repair. After defects are found, human intervention is still required. The degree of automation is limited, and the repair quality cannot be guaranteed to be stable. Summary of the Invention

[0004] In order to solve the problems existing in the prior art mentioned above, the present invention provides an autonomous detection and repair method and system for insulated overhead lines that integrates "detection-judgment-repair-verification".

[0005] The technical solution adopted by this invention to solve its technical problem is: a method for autonomous detection and repair of damage to insulated overhead lines, comprising the following steps:

[0006] S1. Timestamp the data collected by the infrared sensor, vision sensor and vibration sensor, and map the infrared pixel coordinates, image pixel coordinates and partial discharge source coordinates to the basic coordinate system of the robot body.

[0007] S2. Calculate the temperature increment within the sampling time window; input the visual image within the time window into the convolutional neural network model to output the crack confidence and mask area; perform frequency domain analysis on the vibration signal within the time window to calculate the vibration anomaly score;

[0008] S3. Calculate the overall confidence level of failure based on temperature increment, crack confidence level, and vibration anomaly score;

[0009] S4. If the overall confidence level of the damage is not less than the damage threshold, it is determined to be a real damage, triggering a shutdown and 3D measurement, and then performing repair.

[0010] Furthermore, the formula for calculating the overall confidence level of damage in step S3 is as follows:

[0011]

[0012] In the formula, The overall confidence level for damage; For thermal-visual coupling function, For temperature increment, Crack confidence level; Scoring for vibration anomalies; This is a spatial consistency penalty term. The three-dimensional distance between the geometric center coordinates of the mask region output by the convolutional neural network model and the spatial coordinates of the partial discharge source located by the vibration sensor; , and For weights.

[0013] Furthermore, the weights , and Dynamic adjustments are made based on the physical stage of damage, as detailed below:

[0014] When the damage is in the initial carbonization stage, then ;

[0015] When the damage is in the intermediate crack stage, then ;

[0016] When the damage is at the severe breakdown stage, then .

[0017] Furthermore, in step S2, the visual image is preprocessed before being input into the convolutional neural network model. The preprocessing process is as follows:

[0018] A polarization filter is added to the vision sensor, and the angle of the polarization filter is rotated to acquire multiple frames of visual images. Then, the polarization degree map is calculated to separate the specular reflection and diffuse reflection components of the insulating skin surface. Finally, adaptive histogram equalization is performed on the diffuse reflection component.

[0019] Furthermore, in step S2, the convolutional neural network model includes a backbone network, a neck, and a detection head. The backbone network extracts features from the visual image and inputs them into the neck. The neck performs multi-scale fusion of the features and inputs them into the detection head. The detection head outputs the crack confidence and the mask area.

[0020] Furthermore, the backbone network includes a CHB module, a deformable convolutional layer, and an SPPF module connected in sequence. Several CHB modules are configured and connected sequentially. The neck region includes a top-down feature fusion path module, a bottom-up feature fusion path module, a C3 module, and a coordinate attention module. The top-down feature fusion path module, the bottom-up feature fusion path module, and the C3 module are connected sequentially. Multiple C3 modules are configured and connected sequentially. The coordinate attention modules are embedded at the intersection nodes of the top-down feature fusion path module and the bottom-up feature fusion path module.

[0021] Furthermore, the calculation process for the vibration anomaly score in step S2 is as follows:

[0022] S21. The collected vibration signal is divided into frames, the short-time energy envelope of each frame of vibration signal is calculated, and a Fourier fast transform is performed on each frame of vibration signal to output the spectral amplitude.

[0023] S22. Divide the spectral amplitude of each frame by the short-time energy envelope of the corresponding frame to achieve energy normalization;

[0024] S23. Extract the power frequency voltage phase, normalize the energy and synchronize the spectrum amplitude with the power frequency voltage phase, and introduce a phase compensation factor based on the robot's walking speed to correct the extracted power frequency voltage phase in real time, and construct an anti-drift phase resolution map.

[0025] S24. Based on the phase-resolved spectrum, calculate the vibration anomaly score using the following formula:

[0026]

[0027] In the formula, for Normalization function; The energy entropy of the characteristic frequency band of partial discharge; Phase concentration; The discharge repetition rate; , and For weights.

[0028] Furthermore, the 3D measurement process in step S4 is as follows:

[0029] A 3D vision sensor scans the damaged area, acquires point cloud data, and calculates the length and width of the damaged area as well as the diameter of the wire.

[0030] Furthermore, it also includes step S5. The robot body continues to walk a small distance and executes steps S1 to S4 again. If the temperature increment is lower than the temperature threshold and the overall confidence level of the damage is lower than the safety threshold, then the repair is confirmed to be complete.

[0031] An autonomous detection and repair system for damaged insulated overhead power lines includes a robot body and a navigation module mounted on the robot body. Employing the aforementioned autonomous detection and repair method for damaged insulated overhead power lines, the system further includes:

[0032] The data acquisition module, installed on the robot body, is used to collect temperature data, visual images, and vibration signals;

[0033] The repair module, installed on the robot body, is used to repair damaged areas;

[0034] The measurement module, installed on the robot body, is used to measure the length and width of the damaged area and the diameter of the wire;

[0035] The central control module is installed on the robot body or integrated into the terminal. It communicates with the robot body, navigation module, data acquisition module, repair module and measurement module respectively. It is used to store acquired data, synchronize data in time and space, extract independent features, calculate the comprehensive confidence level of damage and identify damage.

[0036] The beneficial effects of this invention are:

[0037] (1) This invention collects multimodal data in real time using infrared sensors, vision sensors, and vibration sensors. Based on the real-time collected multimodal data, it calculates the temperature increment, crack confidence, and vibration anomaly score within the sampling time window. Then, based on the temperature increment, crack confidence, and vibration anomaly score, it calculates the overall damage confidence. According to the relationship between the overall damage confidence and the damage threshold, it realizes autonomous detection and judgment of damage. Finally, it triggers shutdown and 3D measurement. The 3D measurement obtains the size information of the damaged area and performs repair, realizing autonomous repair of damage. Compared with traditional inspection methods, the detection accuracy is significantly improved, the repair quality is stable, and quantitative size information can be obtained. Compared with existing inspection robots, it can both detect and repair, and the whole process is automated with stable repair quality.

[0038] (2) The present invention timestamps the data collected by infrared sensors, vision sensors and vibration sensors, and maps the infrared pixel coordinates, image pixel coordinates and partial discharge source coordinates to the basic coordinate system of the robot body, so as to meet the multimodal data alignment requirements of "same spatial point, same sampling time window" under dynamic inspection, thereby avoiding the problem of spatiotemporal mismatch between infrared hotspots, visual images and vibration sources caused by shaking when the robot body walks on overhead lines.

[0039] (3) The present invention adds a polarization filter to the vision sensor, rotates the angle of the polarization filter to obtain multiple frames of visual images, then calculates the polarization degree map, separates the specular reflection and diffuse reflection components of the insulating skin surface, and finally performs adaptive histogram equalization on the diffuse reflection component, so that the visual image input to the convolutional neural network model is processed by polarized light de-reflection, completely eliminating sunlight interference, thereby solving the problem of strong reflection on the insulating skin surface.

[0040] (4) The present invention introduces a deformable convolutional layer in the shallow layer of the backbone network. Since the shallow layer retains the texture details of carbon traces, cracks and breakdowns on the surface of the insulation, the deformable convolutional layer can adaptively adjust the sampling points here, thereby accurately capturing the discontinuous jump distribution of carbonization traces, thus avoiding the standard convolution from misjudging them as noise and improving the recall rate.

[0041] (5) The present invention introduces a coordinate attention module in the neck area. The coordinate attention module encodes the horizontal and vertical position information of the feature map respectively, strengthens the response weight of the longitudinal continuous crack, and suppresses background interference such as transverse binding lines and bird spikes, thereby improving the crack recognition accuracy.

[0042] (6) This invention constructs an anti-drift phase-resolved spectrum, analyzes the discharge phase distribution pattern of the phase-resolved spectrum, and then distinguishes between partial discharge and corona interference, thereby solving the problem of partial discharge signal phase reference drift when the robot body walks on overhead lines.

[0043] (7) The present invention re-inspects and verifies the damaged area after repair, thereby further ensuring the stability of the repair quality. Attached Figure Description

[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0045] Figure 1 This is a flowchart of the autonomous detection and repair method for damaged insulated overhead lines in this invention.

[0046] Figure 2 This is a diagram of the backbone network architecture in this invention.

[0047] Figure 3 This is a structural diagram of the neck in this invention.

[0048] Figure 4 This is a framework diagram of the autonomous detection and repair system for damaged insulated overhead lines in this invention.

[0049] In the diagram: 100, robot body; 200, navigation module; 300, data acquisition module; 400, repair module; 500, measurement module; 600, central control module. Detailed Implementation

[0050] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0051] Example 1

[0052] like Figure 1 As shown, a method for autonomous detection and repair of damage to insulated overhead lines includes the following steps:

[0053] S1. Timestamp the data collected by the infrared sensor, vision sensor and vibration sensor, and map the infrared pixel coordinates, image pixel coordinates and partial discharge source coordinates to the basic coordinate system of the robot body 100.

[0054] The data collected by the infrared sensor, vision sensor and vibration sensor are timestamped, and the infrared pixel coordinates, image pixel coordinates and partial discharge source coordinates are uniformly mapped to the basic coordinate system of the robot body 100. This satisfies the multimodal data alignment requirements of "same spatial point, same sampling time window" under dynamic inspection, thereby avoiding the spatiotemporal mismatch problem of infrared hotspots, visual images and vibration sources caused by shaking when the robot body 100 walks on overhead lines.

[0055] Specifically, an infrared sensor collects temperature data, a vision sensor captures visual images, and a vibration sensor collects vibration signals. The PTP (Precise Time Protocol) is used to timestamp the data collected by the infrared, vision, and vibration sensors, ensuring accurate time synchronization. Through pre-calibration, the infrared pixel coordinates, image pixel coordinates, and partial discharge source coordinates are uniformly mapped to the basic coordinate system of the robot body 100.

[0056] S2. Calculate the temperature increment within the sampling time window; input the visual image within the time window into the convolutional neural network model, and output the crack confidence and mask area (i.e., the damaged area); perform frequency domain analysis on the vibration signal within the time window, and calculate the vibration anomaly score.

[0057] Specifically, the convolutional neural network model adopts existing technology. The convolutional neural network model includes a backbone network, a neck and a detection head. The backbone network extracts features from the visual image and inputs them into the neck. The neck performs multi-scale fusion of the features and inputs them into the detection head. The detection head outputs the crack confidence and the mask area.

[0058] In one feasible embodiment, the visual image is preprocessed before being input into the convolutional neural network model. The preprocessing process is as follows:

[0059] A polarization filter is added to the vision sensor, and the angle of the polarization filter is rotated to acquire multiple frames of visual images. Then, the polarization degree map is calculated to separate the specular reflection and diffuse reflection components of the insulating skin surface. Finally, adaptive histogram equalization is performed on the diffuse reflection component.

[0060] Because the surface of the insulation is smooth, natural light will cause specular reflection, which may cause general convolutional neural network models to misjudge the reflection as cracks or obscure real microcracks. The preprocessing described above allows the visual image input to the convolutional neural network model to undergo polarized light de-reflection processing, completely eliminating sunlight interference and thus solving the problem of strong reflection on the insulation surface.

[0061] In one feasible embodiment, such as Figure 2 As shown, the backbone network includes CHB modules, deformable convolutional layers, and SPPF modules connected in sequence. Multiple CHB modules are configured and connected in sequence.

[0062] Multiple CHB modules are stacked sequentially in the shallow layer to extract detailed features such as edges and textures on the surface of the insulation skin. They are then connected to deformable convolutional layers, and finally aggregated through multi-scale pooling using the SPPF module before being output to the neck.

[0063] The carbonized channels formed by subsurface discharge in insulation often project as discontinuous, intermittent dot-like patterns on the surface. The fixed receptive field of standard convolution will misidentify these discontinuous points as noise. This application introduces deformable convolutional layers in the shallow layers of the backbone network. Since the shallow layers retain the texture details of carbon traces, cracks, and breakdowns on the insulation surface, the deformable convolutional layers can adaptively adjust the sampling points here, thereby accurately capturing the discontinuous, jump-like distribution of carbonized traces, thus avoiding the misidentification of them as noise by standard convolution and improving recall.

[0064] In some embodiments, such as Figure 3 As shown, the neck adopts a bidirectional feature pyramid structure, including a top-down feature fusion path module, a bottom-up feature fusion path module, a C3 module, and a coordinate attention module. The top-down feature fusion path module, the bottom-up feature fusion path module, and the C3 module are connected in sequence. Multiple C3 modules are set and connected in sequence. The coordinate attention module is embedded at the intersection node of the top-down feature fusion path module and the bottom-up feature fusion path module.

[0065] The top-down feature fusion path propagates deep semantic features upwards layer by layer and fuses them with shallow features at the corresponding scale. The bottom-up feature fusion path propagates the enhanced shallow detail features downwards, enabling full interaction among multi-scale features. Coordinate attention modules are embedded at the intersection of the top-down and bottom-up feature fusion path modules, respectively encoding horizontal and vertical positional information in the fused multi-scale feature map. This strengthens the response weights of longitudinal continuous cracks while suppressing background interference such as transverse binding lines and bird spikes. The fused features are refined through multiple C3 modules and then input into the detection head.

[0066] The background of overhead power lines contains numerous horizontally distributed binding wires and bird spikes, which resemble the texture of longitudinal cracks. This application introduces a coordinate attention module in the neck area, which encodes the horizontal and vertical position information of the feature map, thereby strengthening the response weight of longitudinal continuous cracks and suppressing background interference such as horizontal binding wires and bird spikes, thus improving the crack recognition accuracy.

[0067] The following example, using two intersecting nodes, briefly illustrates how the coordinate attention module is embedded at the intersection of the top-down feature fusion path module and the bottom-up feature fusion path module.

[0068] The backbone network outputs multi-scale feature maps. At the first-layer intersection node of the top-down feature fusion path module: high-level semantic features are upsampled and added to corresponding scale shallow features, then transmitted to the coordinate attention module, finally outputting fused feature 1. At the second-layer intersection node: fused feature 1 is upsampled and added to corresponding scale shallow features, then transmitted to the coordinate attention module, finally outputting fused feature 2. At the first-layer intersection node of the bottom-up feature fusion path module: fused feature 2 is downsampled and added to corresponding scale mid-level features, then transmitted to the coordinate attention module, finally outputting enhanced feature 1. At the second-layer intersection node: enhanced feature 1 is downsampled and added to corresponding scale deep features, then transmitted to the coordinate attention module, finally outputting enhanced feature 2.

[0069] S3. Based on temperature increment, crack confidence level, and vibration anomaly score, calculate the overall confidence level of failure using the following formula:

[0070]

[0071] In the formula, The overall confidence level for damage; For thermal-visual coupling function, For temperature increment, Crack confidence level; Scoring for vibration anomalies; This is a spatial consistency penalty term. The three-dimensional distance between the geometric center coordinates of the mask region output by the convolutional neural network model and the spatial coordinates of the partial discharge source located by the vibration sensor; , and For weights.

[0072] Specifically, thermal-visual coupling function A physical contradiction verification logic is employed, combining temperature and visual characteristics. If a certain area experiences a temperature rise and exhibits circular diffusion, but the visually perceived crack appears linear, it is determined to be due to concentrated sunlight rather than damage. In this case, the thermal-visual coupling function... The value decreases. Because the robot's 100° movement causes changes in the sensor's field of view, and the visual positioning error varies with the guide wire sag, a spatial consistency penalty term is introduced into the overall confidence level of the damage assessment. Spatial consistency penalty A conductor sag nonlinearity correction factor was introduced. Weights , and It is not obtained from machine learning training, but is based on a preset physical stage of damage.

[0073] In one feasible embodiment, weights , and Dynamic adjustments are made based on the physical stage of damage, as detailed below:

[0074] When the damage is in the early carbonization stage (no obvious cracks on the insulation surface, but discontinuous carbon marks appear under the insulation), then It focuses on vibration to capture early partial discharge signals;

[0075] When the damage is in the intermediate crack stage (the carbonized channels are connected, and linear microcracks appear on the surface of the insulation), then Both temperature and vibration are important;

[0076] When the damage is at the severe breakdown stage (through-insulation damage with obvious cracks), then It focuses on temperature, prioritizing judgment based on temperature increments to quickly locate high-risk damage.

[0077] Among them, weight It always has the lowest weight and is only used for spatial consistency verification; it does not participate in the main decision-making process.

[0078] This application can cover three stages of damage detection, with different weights assigned to each stage, which avoids both missing early minor damage and misjudging high-risk damage in the later stages.

[0079] S4. If the overall confidence level of the damage is not less than the damage threshold, it is determined to be a real damage, triggering a shutdown and 3D measurement, and then performing repair.

[0080] Specifically, the 3D measurement process is as follows: a 3D vision sensor scans the damaged area, acquires point cloud data, and calculates the length and width of the damaged area as well as the diameter of the wire.

[0081] Multimodal data is collected in real time using infrared sensors, vision sensors, and vibration sensors. Based on the collected multimodal data, the temperature increment, crack confidence level, and vibration anomaly score within the sampling time window are calculated. Then, based on the temperature increment, crack confidence level, and vibration anomaly score, the overall damage confidence level is calculated. According to the relationship between the overall damage confidence level and the damage threshold, autonomous damage detection and judgment are achieved. Finally, a shutdown and 3D measurement are triggered. The 3D measurement obtains the dimensional information of the damaged area and performs repair, achieving autonomous damage repair. Compared with traditional inspection methods, the detection accuracy is significantly improved, the repair quality is stable, and quantitative dimensional information can be obtained. Compared with existing inspection robots, it can both detect and repair, with full automation and stable repair quality.

[0082] The calculation process for the vibration anomaly score in step S2 is as follows:

[0083] S21. The collected vibration signal is divided into frames, the short-time energy envelope of each frame is calculated to reflect the changing trend of contact pressure, and Fourier fast transform is performed on each frame of vibration signal to output the spectral amplitude.

[0084] Specifically, vibration signals are collected, and filters are used to retain the sensitive frequency band of partial discharge (such as 40kHz to 200kHz) while filtering out irrelevant frequency bands such as power frequency interference and mechanical vibration.

[0085] S22. Divide the spectral amplitude of each frame by the short-time energy envelope of the corresponding frame to achieve energy normalization, eliminate the influence of contact impedance changes and environmental noise, and can truly reflect the energy intensity of partial discharge and ensure the comparability of partial discharge energy.

[0086] S23. Extract the power frequency voltage phase, normalize the energy, and synchronize the spectral amplitude with the power frequency voltage phase. Then, introduce the spectral amplitude based on the robot's walking speed of 100. Phase compensation factor The extracted power frequency voltage phase is corrected in real time to construct a drift-resistant phase-resolved spectrum. ,in, To correct the phase of the power frequency voltage, This represents the spectral amplitude after energy normalization. This represents the number of discharges for the corresponding phase.

[0087] Specifically, phase compensation factor , The phase drift coefficient (unit: rad·s / m) represents the phase shift of the power frequency voltage caused by the robot body 100 traveling a unit length along the guide wire. It can be obtained by fitting the phase drift coefficient through calibration experiments on a test line segment with a known power frequency voltage phase. With walking speed The linear relationship is obtained.

[0088] S24. Based on the phase-resolved spectrum, calculate the vibration anomaly score using the following formula:

[0089]

[0090] In the formula, for Normalization function, ; The energy entropy of the characteristic frequency band of partial discharge; Phase concentration; The discharge repetition rate; , and For weights.

[0091] Specifically, based on vibration anomaly scoring The numerical range can determine the physical stage of damage, guiding the selection of subsequent repair strategies. For example, minor carbonization only requires local application of insulating varnish, while severe breakdown requires the installation of insulating sleeves. Weighting , and The coefficients were obtained by offline calibration of the partial discharge sample library and are fixed empirical coefficients.

[0092] Existing partial discharge detection methods rely on static sensors. When the robot body 100 moves along overhead power lines, changes in contact impedance cause phase reference drift. This application addresses this issue by constructing an anti-drift phase-resolved spectrum and analyzing the discharge phase distribution pattern within the spectrum to distinguish between partial discharge and corona interference. This solves the problem of partial discharge signal phase reference drift when the robot body 100 moves along overhead power lines.

[0093] In some embodiments, step S5 is also included. The robot body 100 continues to walk a small distance and executes steps S1 to S4 again. If the temperature increment is lower than the temperature threshold and the overall confidence level of the damage is lower than the safety threshold, then the repair is confirmed to be complete.

[0094] Specifically, the preferred temperature threshold is 1℃. The preferred safety threshold is the background noise level. During the re-inspection, if the overall confidence level of the damage is higher than the damage threshold, it indicates that the repair has failed and the damage still exists, and the repair process is repeated. If the overall confidence level of the damage is between the safety threshold and the damage threshold, it is determined to be a residual hidden danger after repair, but it does not constitute high-risk damage. The location information is recorded and marked as "to be observed," and it is included in the next inspection plan for key attention. The robot body will not repeat the repair and will continue to move forward to avoid over-repairing damaged wires. The safety threshold is lower than the damage threshold, with a certain margin in between, to prevent boundary jitter from causing repeated start-stops.

[0095] The repaired damaged area is re-inspected to further ensure the stability of the repair quality.

[0096] Example 2

[0097] like Figure 4 As shown, an autonomous detection and repair system for damaged insulated overhead lines adopts the autonomous detection and repair method for damaged insulated overhead lines in Embodiment 1. It includes a robot body 100 and a navigation module 200, a data acquisition module 300, a repair module 400, a measurement module 500, and a central control module 600 installed on the robot body 100. The navigation module 200 guides the robot body 100's movement along the overhead line. The data acquisition module 300 collects temperature data, visual images, and vibration signals. The repair module 400 repairs damaged areas. The measurement module 500 measures the length and width of the damaged area and the diameter of the conductor. The central control module 600 is communicatively connected to the robot body 100, navigation module 200, data acquisition module 300, repair module 400, and measurement module 500, and is used for storing collected data, data spatiotemporal synchronization, independent feature extraction, damage comprehensive confidence calculation, damage identification, and re-inspection verification.

[0098] Specifically, the data acquisition module 300 includes an infrared sensor, a vision sensor, and a vibration sensor. The infrared sensor is preferably an infrared thermal imager, and the vision sensor is preferably an industrial camera. The measurement module 500 uses a 3D vision sensor.

[0099] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for autonomous detection and repair of damage to insulated overhead lines, characterized in that, Includes the following steps: S1. Timestamp the data collected by the infrared sensor, vision sensor and vibration sensor, and map the infrared pixel coordinates, image pixel coordinates and partial discharge source coordinates to the basic coordinate system of the robot body (100); S2. Calculate the temperature increment within the sampling time window; input the visual image within the time window into the convolutional neural network model to output the crack confidence and mask area; perform frequency domain analysis on the vibration signal within the time window to calculate the vibration anomaly score; S3. Calculate the overall confidence level of failure based on temperature increment, crack confidence level, and vibration anomaly score; S4. If the overall confidence level of the damage is not less than the damage threshold, it is judged as a real damage, triggering a shutdown and 3D measurement, and then performing repair. The calculation process for the vibration anomaly score is as follows: S21. The collected vibration signal is divided into frames, the short-time energy envelope of each frame of vibration signal is calculated, and a Fourier fast transform is performed on each frame of vibration signal to output the spectral amplitude. S22. Divide the spectral amplitude of each frame by the short-time energy envelope of the corresponding frame to achieve energy normalization; S23. Extract the power frequency voltage phase, normalize the energy and synchronize the spectrum amplitude with the power frequency voltage phase, and introduce a phase compensation factor based on the walking speed of the robot body (100) to correct the extracted power frequency voltage phase in real time, and construct an anti-drift phase resolution spectrum. S24. Based on the phase-resolved spectrum, calculate the vibration anomaly score using the following formula: In the formula, for Normalization function; The energy entropy of the characteristic frequency band of partial discharge; Phase concentration; The discharge repetition rate; , and As weight; The formula for calculating the overall confidence level of damage is as follows: In the formula, The overall confidence level for damage; For thermal-visual coupling function, For temperature increment, Crack confidence level; Scoring for vibration anomalies; This is a spatial consistency penalty term. The three-dimensional distance between the geometric center coordinates of the mask region output by the convolutional neural network model and the spatial coordinates of the partial discharge source located by the vibration sensor; , and For weights.

2. The method for autonomous detection and repair of damage to insulated overhead lines according to claim 1, characterized in that, The weight , and Dynamic adjustments are made based on the physical stage of damage, as detailed below: When the damage is in the initial carbonization stage, then ; When the damage is in the intermediate crack stage, then ; When the damage is at the severe breakdown stage, then .

3. The method for autonomous detection and repair of damage to insulated overhead lines according to claim 1, characterized in that, In step S2, the visual image is preprocessed before being input into the convolutional neural network model. The preprocessing process is as follows: A polarization filter is added to the vision sensor, and the angle of the polarization filter is rotated to acquire multiple frames of visual images. Then, the polarization degree map is calculated to separate the specular reflection and diffuse reflection components of the insulating skin surface. Finally, adaptive histogram equalization is performed on the diffuse reflection component.

4. The method for autonomous detection and repair of damage to insulated overhead lines according to claim 1, characterized in that, In step S2, the convolutional neural network model includes a backbone network, a neck, and a detection head. The backbone network extracts features from the visual image and inputs them into the neck. The neck performs multi-scale fusion of the features and inputs them into the detection head. The detection head outputs the crack confidence score and the mask area.

5. The method for autonomous detection and repair of damage to insulated overhead lines according to claim 4, characterized in that, The backbone network includes a CHB module, a deformable convolutional layer, and an SPPF module connected in sequence. Several CHB modules are configured and connected sequentially. The neck network includes a top-down feature fusion path module, a bottom-up feature fusion path module, a C3 module, and a coordinate attention module. The top-down feature fusion path module, the bottom-up feature fusion path module, and the C3 module are connected sequentially. Multiple C3 modules are configured and connected sequentially. The coordinate attention modules are embedded at the intersection nodes of the top-down feature fusion path modules and the bottom-up feature fusion path modules.

6. The method for autonomous detection and repair of damage to insulated overhead lines according to claim 1, characterized in that, The 3D measurement process in step S4 is as follows: The 3D vision sensor scans the damaged area, acquires point cloud data, and calculates the length and width of the damaged area as well as the diameter of the wire.

7. The method for autonomous detection and repair of damage to insulated overhead lines according to claim 1, characterized in that, It also includes step S5. The robot body (100) continues to walk a small distance and executes steps S1 to S4 again. If the temperature increment is lower than the temperature threshold and the overall confidence level of the damage is lower than the safety threshold, then the repair is confirmed to be complete.

8. A system for autonomous detection and repair of damaged insulated overhead lines, comprising a robot body (100) and a navigation module (200) mounted on the robot body (100), characterized in that, The method for autonomous detection and repair of damage to insulated overhead lines according to any one of claims 1 to 7 further includes: A data acquisition module (300) is installed on the robot body (100) and is used to acquire temperature data, visual images and vibration signals; A repair module (400) is installed on the robot body (100) and is used to repair the damaged area; A measurement module (500), mounted on the robot body (100), is used to measure the length and width of the damaged area and the diameter of the wire; The central control module (600) is installed on the robot body (100) or integrated into the terminal. It is connected to the robot body (100), navigation module (200), data acquisition module (300), repair module (400) and measurement module (500) respectively. It is used to store acquired data, synchronize data in time and space, extract independent features, calculate the comprehensive confidence level of damage and identify damage.

Citation Information

Patent Citations

  • Container surface damage detection method and device based on machine vision

    CN120525815A

  • Pipeline damage detection method, device, equipment and medium

    CN122544264A