Dynamic path planning power distribution inspection robot based on laser radar

Abstract

The invention discloses a dynamic path planning power distribution inspection robot based on laser radar, which relates to the technical field of power distribution inspection and comprises a mobile carrier, a laser radar sensing module, an environment modeling unit, a dynamic path planning unit, an inspection execution unit and a remote communication unit. The laser radar sensing module collects three-dimensional environment data of a power distribution scene in real time, the environment modeling unit constructs a dynamic environment map based on the collected data, and the dynamic path planning unit generates and dynamically adjusts an inspection path through an improved dynamic planning algorithm in combination with power distribution inspection task requirements and real-time environment information. The inspection execution unit completes line defect detection and data acquisition, and the remote communication unit achieves data interaction with a background monitoring center. The problems that a traditional power distribution inspection robot is poor in path adaptability and insufficient in obstacle avoidance capacity in a complex environment are solved, the inspection efficiency and operation safety are improved, and the robot is suitable for complex and changeable power distribution inspection scenes.

Description

Technical Field

[0001] This invention relates to the field of power distribution inspection equipment technology, and in particular to a power distribution inspection robot based on dynamic path planning using lidar. Background Technology

[0002] The power distribution system is a crucial link in power transmission, and its operational status directly affects the stability of power supply. Traditional power distribution inspections mainly rely on manual labor, which suffers from high labor intensity, low inspection efficiency, and high operational risks. With the development of robotics technology, power distribution inspection robots are gradually being applied in practical operations, but existing products still have shortcomings:

[0003] 1. Path planning is mostly static planning, with fixed paths set in advance. It cannot cope with dynamic obstacles in power distribution scenarios (such as temporary construction equipment, fallen tree branches), and is prone to path blockage or collision risks.

[0004] 2. Environmental perception relies heavily on a single visual sensor, which is greatly affected by lighting and weather conditions, and has insufficient accuracy in recognizing complex environments.

[0005] 3. The route planning did not fully take into account the characteristics of the power distribution inspection task, and the inspection priority of key equipment was not given enough consideration, resulting in weak inspection focus.

[0006] LiDAR has advantages such as high ranging accuracy, strong anti-interference ability, and real-time acquisition of three-dimensional environmental information. Combining it with dynamic path planning algorithms can effectively solve the adaptability problem of existing inspection robots and improve the intelligence level of inspection operations.

[0007] Therefore, it is necessary to propose a power distribution inspection robot based on dynamic path planning using lidar to address the above problems. Summary of the Invention

[0008] To address the shortcomings of the existing technologies, the present invention aims to provide a power distribution inspection robot based on dynamic path planning using lidar.

[0009] A power distribution inspection robot based on lidar and dynamic path planning includes:

[0010] Mobile carrier, used to carry various functional modules and enable movement during the inspection process;

[0011] A lidar sensing module is installed on the top of the mobile carrier to scan the power distribution scene in real time and collect three-dimensional point cloud data, which includes obstacle locations, power distribution equipment outlines and line distribution information.

[0012] An environmental modeling unit, electrically connected to the lidar sensing module, is used to perform noise reduction and registration processing on the three-dimensional point cloud data to construct a dynamically updated power distribution scene grid map.

[0013] The dynamic path planning unit is electrically connected to the environment modeling unit and the inspection task input module, respectively. It has a built-in improved D*Lite algorithm to dynamically generate the optimal inspection path based on the inspection task path start point, end point, key inspection points and real-time grid map, and adjust the path in real time when the environment changes.

[0014] The inspection execution unit includes a high-definition camera, an infrared thermal imager, and a data acquisition card, which are used to acquire images, detect temperatures, and identify defects in power distribution equipment along a planned path.

[0015] The remote communication unit is electrically connected to the dynamic path planning unit and the inspection execution unit respectively, and is used to upload inspection data to the background monitoring center and receive background control commands.

[0016] The central control unit is electrically connected to the mobile carrier, the lidar sensing module, the environmental modeling unit, the dynamic path planning unit, the inspection execution unit, and the remote communication unit, respectively, and is used to coordinate the collaborative work of each module.

[0017] The LiDAR sensing module has a scanning range of 360°, a ranging accuracy of ≤±2cm, a scanning frequency of 10-20Hz, and supports real-time output of 3D point cloud data.

[0018] The processing flow of the environment modeling unit includes:

[0019] S1. Perform pass-through filtering on the raw point cloud data collected by the lidar to remove noise and eliminate distance anomalies;

[0020] S2. Point cloud registration is performed using the iterative nearest point algorithm to achieve spatial fusion of multi-frame data;

[0021] S3. Based on the occupancygrid algorithm, the fused point cloud data is converted into a raster map with a raster resolution of 0.1m × 0.1m;

[0022] S4. Update the grid map every 0.5 seconds and mark the dynamic obstacle area.

[0023] The optimizations of the improved D*Lite algorithm include:

[0024] S11. Introduce priority weights for power distribution equipment and set higher access weights for key inspection points than for ordinary areas.

[0025] S12. Use heuristic function optimization and combine it with power distribution line routing constraints to reduce path search redundancy;

[0026] S13. Add a path smoothing module to eliminate path inflection points through Bézier curve fitting and adapt to the motion characteristics of the mobile carrier.

[0027] The steps of the iterative nearest point algorithm are as follows:

[0028] S21. Initialization, given an initial rotation matrix and the initial translation vector ;

[0029] S22. Find corresponding point pairs for each point in the source point cloud. Find the nearest point in the target point cloud. ; ,

[0030] S23. Calculate the optimal rotation matrix R and translation vector t, based on the corresponding point pairs obtained in step 2. Solve for the objective function The smallest R and t;

[0031] S24. Update the source point cloud and calculate the error; transform the source point cloud using R and t obtained in step 3. k is the current iteration number: calculate the error of the current iteration. ;

[0032] S25. Determine convergence; if... (Error change is less than the threshold) or (If the maximum number of iterations is reached), terminate the iteration and output the final transformation matrix R and t); otherwise, return to step 2 and repeat the iteration.

[0033] in For source point cloud, For three-dimensional coordinate points: For the target point cloud, These are three-dimensional coordinate points.

[0034] S23 further states:

[0035] S31. Calculate the centroid, the centroid of the source point cloud: ;

[0036] Target point cloud centroid: ;

[0037] S32. Centroid removal process: The point cloud is translated to a coordinate system with the centroid as the origin to eliminate the influence of translation on rotation calculations.

[0038] Source point centroid removal: ;

[0039] Target point centroid: ;

[0040] S33. Construct the covariance matrix and solve for the rotation matrix R;

[0041] Define the covariance matrix H, Perform singular value decomposition on H. Where U and V are orthogonal matrices, It is a diagonal matrix; the optimal rotation matrix is: ,

[0042] like Correction needed: ,in Calculate the translation vector t, based on the rotation matrix and the centroid, the translation vector is: .

[0043] The steps of the occupancygrid algorithm are as follows:

[0044] S41. Raster map initialization, setting raster resolution. Define the map area and generate

[0045] Raster matrix; initialize the log-odds of each raster cell. ;

[0046] S42. LiDAR data preprocessing: Input single-frame 3D point cloud data from the LiDAR. in ,for

[0047] Obstacle point coordinates are obtained by projecting the 3D point cloud onto a 2D plane. ,

[0048] Facilitates raster mapping;

[0049] S43. Determine the correlation between the grid and the observation.

[0050] S44, Grid Probability Update :

[0051] Using S43 Update log odds:

[0052] S45. Generate a raster map and calculate the logarithmic probability of each raster. Convert to occupancy probability: ,

[0053] Set threshold for dividing state:

[0054] like Marked as "occupied" (obstacle, black);

[0055] like If the passable area is marked as "Idle" (passable, white), then it is marked as "Unknown" (gray).

[0056] Preferably, the mobile carrier adopts a tracked chassis, equipped with a differential drive system, with a maximum moving speed of 1.5m / s, a climbing angle of ≤30°, and supports turning on the spot, adapting to complex terrain in power distribution inspection.

[0057] Preferably, the defect identification process of the inspection execution unit includes:

[0058] S51: A high-definition camera captures images of power distribution equipment, while an infrared thermal imager simultaneously collects equipment temperature data.

[0059] S52. The data acquisition card preprocesses the image data, including grayscale conversion and edge detection.

[0060] S53. Use the YOLOv8 algorithm to perform defect detection on the preprocessed image and identify defect types such as insulator breakage, conductor strand breakage, and hardware loosening.

[0061] S54. Based on temperature data, determine whether the equipment has an overheating fault. The abnormal temperature threshold can be set in the background.

[0062] Preferably, the remote communication unit supports 5G / Wi-Fi dual-mode communication, with a data transmission rate of ≥100Mbps and a latency of ≤50ms, enabling real-time uploading of inspection videos, images, temperature data, and robot status information.

[0063] Compared with the prior art, the present invention has the following advantages:

[0064] 1. Based on LiDAR, real-time environmental perception and improved dynamic programming algorithm can quickly respond to environmental changes, avoid dynamic obstacles, and ensure continuous inspection operations;

[0065] 2. The high-precision perception of LiDAR combined with the YOLOv8 defect recognition algorithm improves the accuracy of environmental recognition and defect detection. Dynamic path planning reduces unnecessary movement and improves inspection efficiency by more than 30%.

[0066] 3. Robots replace human workers in complex power distribution scenarios, avoiding the safety risks of working at heights and on live lines, while obstacle avoidance algorithms reduce the risk of collisions to the robot itself.

[0067] 4. Supports setting priority for key inspection points to ensure that critical equipment receives focused inspection and meets the task requirements of power distribution inspection. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the composition of the present invention;

[0069] Figure 2 This is a flowchart of the environmental modeling unit of the present invention;

[0070] Figure 3 This is a flowchart of the improved D*Lite algorithm of the present invention;

[0071] Figure 4 This is a flowchart of the defect identification process of the present invention. Detailed Implementation

[0072] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0073] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0074] like Figure 1 and combined Figures 2 to 4 As shown, a power distribution inspection robot based on lidar dynamic path planning includes:

[0075] Mobile carrier, used to carry various functional modules and enable movement during the inspection process;

[0076] A lidar sensing module is installed on the top of the mobile carrier to scan the power distribution scene in real time and collect three-dimensional point cloud data, which includes obstacle locations, power distribution equipment outlines and line distribution information.

[0077] An environmental modeling unit, electrically connected to the lidar sensing module, is used to perform noise reduction and registration processing on the three-dimensional point cloud data to construct a dynamically updated power distribution scene grid map.

[0078] The dynamic path planning unit is electrically connected to the environment modeling unit and the inspection task input module, respectively. It has a built-in improved D*Lite algorithm to dynamically generate the optimal inspection path based on the inspection task path start point, end point, key inspection points and real-time grid map, and adjust the path in real time when the environment changes.

[0079] The inspection execution unit includes a high-definition camera, an infrared thermal imager, and a data acquisition card, which are used to acquire images, detect temperatures, and identify defects in power distribution equipment along a planned path.

[0080] The remote communication unit is electrically connected to the dynamic path planning unit and the inspection execution unit respectively, and is used to upload inspection data to the background monitoring center and receive background control commands.

[0081] The central control unit is electrically connected to the mobile carrier, the lidar sensing module, the environmental modeling unit, the dynamic path planning unit, the inspection execution unit, and the remote communication unit, respectively, and is used to coordinate the collaborative work of each module.

[0082] The LiDAR sensing module has a scanning range of 360°, a ranging accuracy of ≤±2cm, a scanning frequency of 10-20Hz, and supports real-time output of 3D point cloud data.

[0083] The processing flow of the environment modeling unit includes:

[0084] S1. Perform pass-through filtering on the raw point cloud data collected by the lidar to remove noise and eliminate distance anomalies;

[0085] S2. Point cloud registration is performed using the iterative nearest point algorithm to achieve spatial fusion of multi-frame data;

[0086] S3. Based on the occupancygrid algorithm, the fused point cloud data is converted into a raster map with a raster resolution of 0.1m × 0.1m;

[0087] S4. Update the grid map every 0.5 seconds and mark the dynamic obstacle area.

[0088] The optimizations of the improved D*Lite algorithm include:

[0089] S11. Introduce priority weights for power distribution equipment and set higher access weights for key inspection points than for ordinary areas.

[0090] S12. Use heuristic function optimization and combine it with power distribution line routing constraints to reduce path search redundancy;

[0091] S13. Add a path smoothing module to eliminate path inflection points through Bézier curve fitting and adapt to the motion characteristics of the mobile carrier.

[0092] The steps of the iterative nearest point algorithm are as follows:

[0093] S21. Initialization, given an initial rotation matrix and the initial translation vector ;

[0094] S22. Find corresponding point pairs for each point in the source point cloud. Find the nearest point in the target point cloud. ; ,

[0095] S23. Calculate the optimal rotation matrix R and translation vector t, based on the corresponding point pairs obtained in step 2. Solve for the objective function The smallest R and t;

[0096] S24. Update the source point cloud and calculate the error; transform the source point cloud using R and t obtained in step 3. k is the current iteration number: calculate the error of the current iteration. ;

[0097] S25. Determine convergence; if... (Error change is less than the threshold) or (If the maximum number of iterations is reached), terminate the iteration and output the final transformation matrix R and t); otherwise, return to step 2 and repeat the iteration.

[0098] in For source point cloud, For three-dimensional coordinate points: For the target point cloud, These are three-dimensional coordinate points.

[0099] S23 further states:

[0100] S31. Calculate the centroid, the centroid of the source point cloud: ;

[0101] Target point cloud centroid: ;

[0102] S32. Centroid removal process: The point cloud is translated to a coordinate system with the centroid as the origin to eliminate the influence of translation on rotation calculations.

[0103] Source point centroid removal: ;

[0104] Target point centroid: ;

[0105] S33. Construct the covariance matrix and solve for the rotation matrix R;

[0106] Define the covariance matrix H, Perform singular value decomposition on H. Where U and V are orthogonal matrices, It is a diagonal matrix; the optimal rotation matrix is: ,

[0107] like Correction needed: ,in Calculate the translation vector t, based on the rotation matrix and the centroid, the translation vector is: .

[0108] The steps of the occupancygrid algorithm are as follows:

[0109] S41. Raster map initialization, setting raster resolution. Define the map area and generate

[0110] Raster matrix; initialize the log-odds of each raster cell. ;

[0111] S42. LiDAR data preprocessing: Input single-frame 3D point cloud data from the LiDAR. in ,for

[0112] Obstacle point coordinates are obtained by projecting the 3D point cloud onto a 2D plane. ,

[0113] Facilitates raster mapping;

[0114] S43. Determine the correlation between the grid and the observation.

[0115] S44, Grid Probability Update :

[0116] Using S43 Update log odds:

[0117] S45. Generate a raster map and calculate the logarithmic probability of each raster. Convert to occupancy probability: ,

[0118] Set threshold for dividing state:

[0119] like Marked as "occupied" (obstacle, black);

[0120] like If the passable area is marked as "Idle" (passable, white), then it is marked as "Unknown" (gray).

[0121] Preferably, the mobile carrier adopts a tracked chassis, equipped with a differential drive system, with a maximum moving speed of 1.5m / s, a climbing angle of ≤30°, and supports turning on the spot, adapting to complex terrain in power distribution inspection.

[0122] Preferably, the defect identification process of the inspection execution unit includes:

[0123] S51: A high-definition camera captures images of power distribution equipment, while an infrared thermal imager simultaneously collects equipment temperature data.

[0124] S52. The data acquisition card preprocesses the image data, including grayscale conversion and edge detection.

[0125] S53. Use the YOLOv8 algorithm to perform defect detection on the preprocessed image and identify defect types such as insulator breakage, conductor strand breakage, and hardware loosening.

[0126] S54. Based on temperature data, determine whether the equipment has an overheating fault. The abnormal temperature threshold can be set in the background.

[0127] Preferably, the remote communication unit supports 5G / Wi-Fi dual-mode communication, with a data transmission rate of ≥100Mbps and a latency of ≤50ms, enabling real-time uploading of inspection videos, images, temperature data, and robot status information.

[0128] Compared with the prior art, the present invention has the following advantages:

[0129] 1. Based on LiDAR, real-time environmental perception and improved dynamic programming algorithm can quickly respond to environmental changes, avoid dynamic obstacles, and ensure continuous inspection operations;

[0130] 2. The high-precision perception of LiDAR combined with the YOLOv8 defect recognition algorithm improves the accuracy of environmental recognition and defect detection. Dynamic path planning reduces unnecessary movement and improves inspection efficiency by more than 30%.

[0131] 3. Robots replace human workers in complex power distribution scenarios, avoiding the safety risks of working at heights and on live lines, while obstacle avoidance algorithms reduce the risk of collisions to the robot itself.

[0132] 4. Supports setting priority for key inspection points to ensure that critical equipment receives focused inspection and meets the task requirements of power distribution inspection.

[0133] The server and client communicate via a communication unit to request data. The server is responsible for data management, acquiring, processing, and storing data from each power distribution terminal site. People can access and view information through the server, enabling remote monitoring of power distribution terminal data and maintenance status.

[0134] LiDAR perception module: It adopts 360° omnidirectional LiDAR to ensure no blind spots in the collection of environmental data. The ranging accuracy is ≤±2cm. It can accurately identify the spatial position of power distribution equipment and obstacles, and is not affected by light or weather.

[0135] Environmental modeling unit: Through filtering and noise reduction, point cloud registration and other processing, the original point cloud data is converted into a high-precision raster map and updated in real time to accurately reflect the dynamic changes in the environment.

[0136] Dynamic path planning unit: Based on the improved D*Lite algorithm, combined with the priority of key points and route constraints in the inspection task, the optimal path is generated quickly. When the environment changes (such as the appearance of new obstacles), the path is adjusted in real time to ensure the continuity of the inspection.

[0137] Inspection execution unit: Integrating visual and infrared detection technologies, it achieves rapid defect identification through the YOLOv8 algorithm, and simultaneously collects equipment temperature data to improve the comprehensiveness and accuracy of defect detection.

[0138] Workflow

[0139] Inspection task initialization: Set the inspection area, start point, end point and key inspection points (such as transformers, insulators and circuit breakers) through the background monitoring center, and set the defect identification threshold and temperature anomaly threshold.

[0140] Environmental perception and modeling: After the robot is started, the lidar perception module collects three-dimensional point cloud data in real time, and the environmental modeling unit processes the data to build a dynamic grid map.

[0141] Dynamic path planning: The dynamic path planning unit generates an initial inspection path based on the grid map and inspection task parameters using the improved D*Lite algorithm;

[0142] Inspection operation execution: The mobile carrier moves along the planned path, and the inspection execution unit simultaneously collects equipment images and temperature data to identify defects;

[0143] Dynamic path adjustment: When the lidar detects a new obstacle, the environment modeling unit updates the grid map in real time, and the dynamic path planning unit recalculates the optimal path to avoid the obstacle;

[0144] Data interaction and feedback: Inspection data is uploaded to the backend via a remote communication unit, and the backend can monitor the robot's status in real time and issue control commands.

[0145] Example 1: Outdoor 10kV power distribution line inspection scenario

[0146] For outdoor 10kV power distribution line corridors (including 3km of line, 12 transformers, and 36 sets of insulators), there are dynamic obstacles such as fallen trees, temporary construction fences, and bird nests. Traditional manual inspections are inefficient (approximately 8 hours per inspection) and pose high safety risks. This embodiment uses robots to achieve fully automated and high-precision inspections.

[0147] Mobile carrier, tracked chassis (material: high-strength aluminum alloy, weight: 75kg), differential drive, maximum speed 1.5m / s, climbing angle 30°, supports on-the-spot turning.

[0148] The lidar sensing module is a 16-line lidar (model: Velodyne16) with a 360° scanning range, a ranging range of 0.1-100m, a ranging accuracy of ±2cm, and a scanning frequency of 15Hz.

[0149] Inspection execution unit, 4K high-definition camera (focal length 10-100mm, autofocus); infrared thermal imager (temperature measurement range -20℃~150℃, accuracy ±1℃); data acquisition card (PCI-E interface, sampling frequency 100Hz).

[0150] The central control unit is an NVIDIA Jetson AGXXavier (8-core CPU, 512-core GPU) running Ubuntu 20.04.

[0151] Remote communication unit, 5G / Wi-Fi dual mode (5G band Sub-6GHz, speed ≥100Mbps; Wi-Fi 802.11ac, coverage radius 50m).

[0152] Power module, lithium battery pack (200Ah, 24V), 8 hours of battery life, fast charging time ≤2 hours.

[0153] Environmental modeling unit:

[0154] Straight-through filtering: Removes abnormal points with a distance >50m (effective range for outdoor line inspection);

[0155] ICP registration: 30 iterations, convergence threshold 10. -6 The multi-frame point cloud fusion error is ≤0.5cm;

[0156] OccupancyGrid: Grid resolution 0.1m×0.1m, update frequency 0.5s / time, dynamic obstacle marking delay ≤0.3s.

[0157] Improved D*Lite algorithm:

[0158] Priority weight: Transformer (weight 3) > Insulator (weight 2) > Normal area (weight 1);

[0159] Heuristic function: \(h(n)=0.5×Euclidean distance+0.3×route direction constraint (penalty coefficient reduced by 20% along the route direction)+0.2×priority weight\);

[0160] Path smoothing: Bezier curve fitting, inflection point elimination rate ≥90%, path curvature ≤0.5rad / m (adapted to tracked chassis motion).

[0161] Defect identification:

[0162] YOLOv8 model training: The dataset contains 50,000 images (10 types of defects such as insulator damage and broken conductor strands), with a recognition accuracy of 96.2% and a speed of 30 frames per second;

[0163] Temperature anomaly detection: The threshold is set to >80℃ (can be adjusted in the background), and the overheating fault recognition delay is ≤0.5s.

[0164] 4. Test Results

[0165] Inspection efficiency: The robot can complete a 3km line inspection in just 2 hours, which is 4 times more efficient than manual inspection.

[0166] Obstacle avoidance performance: Successfully avoided temporary construction fences (2m in diameter) and fallen tree branches (3m in length), with a path adjustment response time of 0.2s;

[0167] Defect identification: Accurately identified 3 insulator breaks, 2 loose fittings, and 1 transformer overheating, with no missed or false detections;

[0168] Safety: No collisions or loss of communication throughout the entire process, reducing operational safety risks by 100%.

[0169] Example 2: Indoor power distribution room inspection scenario

[0170] A 110kV power distribution room (500㎡) contains 30 switchgear cabinets, 20 busbar trunking systems, and 15 grounding devices. It presents dynamic obstacles such as tool storage and temporary personnel work, requiring 4 hours per inspection using traditional manual methods, and suffers from limited visibility in densely populated equipment areas. This embodiment utilizes a robot to achieve high-precision inspection in confined spaces.

[0171] Mobile carrier, small tracked chassis (dimensions: 80cm×60cm×50cm, weight: 50kg), differential drive, maximum speed 1.2m / s, minimum turning radius 0m (turning on the spot).

[0172] The lidar sensing module is a 32-line lidar (model: LivoxMid-70) with a 360° scanning range, a ranging range of 0.1-50m, a ranging accuracy of ±1cm, and a scanning frequency of 20Hz.

[0173] Inspection execution unit, high-definition pan-tilt camera (360° rotation, focal length 8-80mm); infrared thermal imager (temperature measurement range -10℃~120℃, accuracy ±0.5℃); data acquisition card (USB3.0 interface, sampling frequency 200Hz).

[0174] The central control unit is an NVIDIA Jetson Xavier NX (6-core CPU, 384-core GPU) running Ubuntu 20.04.

[0175] Remote communication unit: Wi-Fi 6 (802.11ax, speed ≥1Gbps, coverage radius 30m) + Bluetooth 5.0 (near-range backup communication).

[0176] Environmental modeling unit:

[0177] Straight-through filtering: Removes abnormal points with a distance >20m (effective range of power distribution room);

[0178] ICP registration: 25 iterations, convergence threshold 5×10⁻⁶ -7 The multi-frame point cloud fusion error is ≤0.3cm;

[0179] OccupancyGrid: Grid resolution 0.1m×0.1m, update frequency 0.3s / time, dynamic obstacle (such as toolbox, people) marking accuracy 98%.

[0180] Improved D*Lite algorithm:

[0181] Priority weight: Switchgear > Busbar trunking > Grounding device > Channel;

[0182] Heuristic function: \(h(n)=0.4×Euclidean distance+0.4×spatial constraint (penalty coefficient for small areas increases by 30%)+0.2×priority weight\);

[0183] Path smoothing: Bezier curve fitting, path width ≤ 0.8m (adapted to the width of power distribution room passageway).

[0184] Defect identification:

[0185] YOLOv8 model training: The dataset contains 30,000 images (8 types of defects such as abnormal noise from switch cabinets, loose busbars, and poor grounding), with a recognition accuracy of 97.5% and a speed of 40 frames per second;

[0186] Temperature anomaly detection: The threshold is set to >60℃ (normal switch cabinet temperature ≤50℃), and the overheating fault identification delay is ≤0.2s.

[0187] 4. Test Results

[0188] Inspection efficiency: The robot can complete the inspection of a 500㎡ power distribution room in just 1 hour, which is 4 times more efficient than manual inspection.

[0189] Spatial adaptability: Successfully navigated a 1m wide passage and bypassed a toolbox with a 0.5m diameter without getting stuck;

[0190] Defect identification: Accurately identified 2 abnormal noises from the switchgear, 3 overheating issues at busbar joints, and 1 loose grounding bolt, with an identification accuracy >97%.

[0191] Data transmission: Inspection videos and temperature data are uploaded in real time with a delay of ≤20ms and no data loss.

[0192] Power module, lithium battery pack (150Ah, 24V), 6 hours of battery life, fast charging time ≤1.5 hours.

[0193] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic path planning distribution inspection robot based on laser radar, characterized by: The utility model relates to a kind of power distribution inspection robot, including: Mobile carrier, for carrying each functional module and realizing the movement in the process of inspection; Laser radar sensing module, installed on the top of the mobile carrier, for real-time scanning power distribution scene, collecting three-dimensional point cloud data, the three-dimensional point cloud data includes obstacle position, power distribution equipment contour and line distribution information; Environment modeling unit, electrically connected with the laser radar sensing module, for denoising, registration processing to the three-dimensional point cloud data, constructs dynamic update power distribution scene grid map; Dynamic path planning unit, electrically connected with the environment modeling unit and inspection task input module respectively, built-in improved D*Lite algorithm, for dynamically generating optimal inspection path according to inspection task path starting point, end point, inspection key point and real-time grid map, and adjusting path in real time when environment changes; Inspection execution unit, including high-definition camera, infrared thermal imager and data acquisition card, for image acquisition, temperature detection and defect identification to power distribution equipment along the planned path; Remote communication unit, electrically connected with the dynamic path planning unit and inspection execution unit respectively, for uploading inspection data to background monitoring center, and receiving background control instruction; Central control unit, electrically connected with the mobile carrier, laser radar sensing module, environment modeling unit, dynamic path planning unit, inspection execution unit and remote communication unit respectively, for coordinating the collaborative work of each module.

2. The dynamic path planning distribution inspection robot based on laser radar according to claim 1, characterized in that: The scanning range of the laser radar sensing module is 360°, the ranging accuracy is ≤±2cm, the scanning frequency is 10-20Hz, and the real-time output of three-dimensional point cloud data is supported.

3. The dynamic path planning distribution inspection robot based on laser radar according to claim 2, characterized in that: The processing flow of the environment modeling unit includes: S1, straight-through filtering denoising is carried out on the original point cloud data collected by laser radar, and distance abnormal points are removed; S2, point cloud registration is carried out by using iterative closest point algorithm, and spatial fusion of multiple frames of data is realized; S3, the fused point cloud data is converted into grid map based on occupancygrid algorithm, and the grid resolution is 0.1m×0.1m; S4, the grid map is updated once every 0.5s, and dynamic obstacle area is marked.

4. The dynamic path planning distribution inspection robot based on laser radar according to claim 3, characterized in that: The optimization of the improved D*Lite algorithm includes: S11, the priority weight of power distribution equipment is introduced, and the access weight of inspection key point is set to be higher than that of ordinary area; S12, heuristic function optimization is adopted, combined with power line direction constraint, to reduce path search redundancy; S13, a path smoothing processing module is added, and the path inflection point is eliminated through Bezier curve fitting, to adapt to the motion characteristics of the mobile carrier.

5. The dynamic path planning distribution inspection robot based on laser radar according to claim 4, characterized in that: The iterative closest point algorithm step is as follows: S21, initialization, given initial rotation matrix and initial translation vector ; S22, finding a corresponding point pair, for each point in the source point cloud , finding the closest point in the target point cloud ; , S23, calculate the optimal rotation matrix R and translation vector t, based on the corresponding point pairs obtained in step 2 , solve R and t that minimize the objective function ​ S24, update the source point cloud and compute the error; transform the source point cloud with R and t obtained in step 3 k is the current iteration number: compute the error for the current iteration ; S25, judging convergence, if or reaching the maximum iteration number, terminating iteration, outputting the final transformation matrix R and t; otherwise, returning to step 2 and repeating iteration; wherein is a source point cloud, is a three-dimensional coordinate point: is a target point cloud, is a three-dimensional coordinate point.

6. The LiDAR-based dynamic path planning power distribution inspection robot of claim 5, wherein: S23 is further as follows: S31, calculate the centroid, source point cloud centroid: ; Target point cloud centroid: ; S32, centroid removal processing is carried out, and the point cloud is translated to the coordinate system with centroid as origin, to eliminate the influence of translation on rotation calculation: Source point to centroid: ; Target point decentering: ; S33, covariance matrix is constructed and rotation matrix R is solved; define the covariance matrix H, , perform a singular value decomposition of H, , where U and V are orthogonal matrices, is a diagonal matrix; the optimal rotation matrix is: , If , correct: where ; compute the translation vector t based on the rotation matrix and the center of mass, the translation vector being .

7. The dynamic path planning distribution inspection robot based on laser radar according to claim 3, characterized in that: The occupancygrid algorithm step is as follows: S41, initialize the grid map, set the grid resolution , delineate the map range, generate Grid matrix; initialize logit for each grid ; S42, laser radar data preprocessing, input single frame three-dimensional point cloud data of laser radar wherein is Obstacle point coordinates, project the three-dimensional point cloud to a two-dimensional plane to obtain two-dimensional coordinates Convenient for grid mapping; S43, the correlation of grid and observation is determined S44, updating the grid probabilities, : With S43 Update log odds: S45, generating a grid map, converting the log-odds of each grid to an occupancy probability: , Threshold value is set to divide state: If , the marker is occupied; If , the tag is free. Otherwise: mark as unknown.

8. The dynamic path planning distribution inspection robot based on lidar of claim 1, wherein: The mobile carrier adopts tracked chassis, carries differential drive system, the maximum moving speed is 1.5m / s, the climbing angle is ≤30°, supports spot turning, and adapts to complex terrain in power distribution inspection.

9. The dynamic path planning distribution inspection robot based on laser radar of claim 1, wherein: The defect identification process of the inspection execution unit comprises: S51, a high-definition camera collects power distribution equipment images, and an infrared thermal imager synchronously collects equipment temperature data; S52, a data acquisition card pre-processes the image data, including grayscale and edge detection; S53, a YOLOv8 algorithm is used to detect defects in the pre-processed images, and identify insulator damage, wire strand breakage, and loose hardware defect types; S54, combined with the temperature data, determine whether the equipment has an overheating fault, the temperature anomaly threshold can be set through the background.

10. The dynamic path planning distribution inspection robot based on lidar according to claim 1, wherein: The remote communication unit supports 5G / Wi-Fi dual-mode communication, with a data transmission rate of ≥100 Mbps and a delay of ≤50 ms, which can realize real-time uploading of inspection videos, images, temperature data and robot state information.

Images