A global path planning method for power converter valve inspection operation robot

Abstract

The application discloses a kind of global path planning methods for power converter valve inspection operation robot, obtains the spatial point cloud data of converter valve area using three-dimensional structured light camera, constructs two-dimensional environment map containing obstacle inflation information by projection and rasterization;According to the snake shape or spiral shape global theoretical path of full coverage generated by camera scanning step;During inspection execution, the feasibility of path segment is judged by discrete sampling and continuous collision detection;Local obstacle avoidance uses improved A* algorithm, simultaneously executes forward and backward bidirectional search in the limited local search area, and introduces the dynamic heuristic function of coverage potential item and safety score item.Finally, the local path and theoretical path are spliced, and the global feasible inspection path is formed.The application effectively solves the problem that the coverage rate and obstacle avoidance efficiency of traditional method are difficult to consider under complex converter valve hall environment, and is especially suitable for automatic inspection in equipment-intensive scenes such as power converter valve hall.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation and path planning technology for mobile robots, specifically to a global path planning method for a power converter valve inspection robot. Background Technology

[0002] With the rapid development of ultra-high voltage direct current (UHVDC) transmission projects, the power converter valve, as a core device for AC / DC energy conversion, directly impacts the safety and stability of the entire power grid. The converter valve hall has a complex internal structure, dense equipment, and numerous obstacles such as valve towers and cooling water pipes. Traditional manual inspection methods face significant challenges, including low efficiency and high safety risks. Automated inspection using inspection robots has become an industry trend; however, path planning in complex valve hall environments still presents many technical challenges. Existing path planning methods are mainly divided into two levels: global planning and local obstacle avoidance. Regarding global path planning, traditional search algorithms such as Dijkstra's algorithm and A* algorithm primarily solve point-to-point optimal path problems, failing to meet the basic requirement of full coverage for power equipment inspection. While intelligent optimization algorithms such as genetic algorithms and ant colony algorithms can theoretically handle coverage problems, they generally suffer from computational complexity and slow convergence speed, making it difficult to meet the real-time requirements of practical applications. Especially in environments like converter valve halls with dense equipment and complex structures, these algorithms often cannot achieve path optimization while ensuring full coverage.

[0003] In terms of local real-time obstacle avoidance, the Artificial Potential Field (APF) method guides robot movement by constructing a virtual potential field. Although computationally simple, it suffers from local minima and is prone to getting stuck in an equilibrium state in concave obstacle regions. The Dynamic Window (DWA) method handles dynamic obstacles well, but its limited foresight is particularly evident in environments with complex static structures, such as converter valve halls, easily leading to robot deadlock in U-shaped areas. Furthermore, the performance of the DWA algorithm is highly dependent on the fine tuning of multiple weight parameters; improper parameter settings directly affect inspection results and safety.

[0004] While recent deep learning-based methods have demonstrated some environmental adaptability, they require large amounts of training data, necessitate high computational resources, and their reliability and generalization capabilities in real-world industrial scenarios still need to be validated. Given the extremely high safety requirements of the power industry, the engineering application of these methods faces significant challenges.

[0005] Specifically, in the scenario of power converter valve inspection, the following technical bottlenecks exist: First, traditional global planning algorithms cannot achieve full coverage of the detection area, easily leading to missed detections; second, various algorithms struggle to achieve an effective balance in key indicators such as coverage, path efficiency, and obstacle avoidance reliability. Existing methods often fail to simultaneously meet the integrity and safety requirements of inspection tasks.

[0006] Therefore, there is an urgent need in this field to develop an innovative path planning method that can effectively integrate global path planning and local obstacle avoidance, improving the efficiency and reliability of path planning while ensuring the integrity of inspections, and meeting the practical application needs of power inspection robots in complex industrial scenarios. The development of this method is not only significant for promoting intelligent inspection of power equipment, but also provides valuable reference for path planning problems in other industrial sectors. Summary of the Invention

[0007] To address the aforementioned problems, this invention provides a global path planning method for a power converter valve inspection robot. To achieve the above objectives, this invention employs the following technical solution:

[0008] A global path planning method for a power converter valve inspection robot is provided, comprising the following steps:

[0009] S1. Use a 3D structured light camera to acquire 3D point cloud data of the converter valve area, and project the acquired 3D point cloud data onto the plane where the detection area is located; construct a 2D raster map based on the projected data, and perform safety expansion processing on environmental obstacles to obtain a 2D raster map containing obstacle information and used for path planning.

[0010] S2. Generate a serpentine or spiral global theoretical path based on the horizontal and vertical scanning step sizes set according to the camera scanning window size;

[0011] S3. Perform validity checks on the path segments between adjacent path points along the global theoretical path, and use discrete sampling and continuous collision detection to determine whether the path segments overlap with the expanded obstacles in the two-dimensional grid map; if the path segment is detected to be impassable, determine the starting point and target point of local obstacle avoidance;

[0012] S4. Construct a local search region centered on the region where the impassable path segment is located; simultaneously perform forward and backward A* search processes within the local search region; constrain the expandable neighborhood based on the global theoretical path direction, and perform continuous collision detection on candidate nodes; generate local obstacle avoidance paths using a dynamic heuristic function that includes coverage potential and safety score terms;

[0013] The expression for the dynamic heuristic function is:

[0014] h(n) = βU(n) - γS(n);

[0015] Where β represents the weighting coefficient of the coverage potential term; γ represents the weighting coefficient of the safety score term; U(n) represents the coverage potential term, which measures whether there are areas around the node that have not yet been scanned; S(n) represents the safety score term, which reflects the relative safety of the node and the set of expanding obstacles.

[0016] S5. Replace the corresponding path segment in the global theoretical path with the local obstacle avoidance path to form a new global theoretical path; the robot performs subsequent inspection tasks along the updated global theoretical path until the entire area is covered by inspection.

[0017] Further, step S1 specifically includes:

[0018] S11. Use a 3D structured light camera to acquire 3D point cloud data of obstacles in the converter valve area, and store it as point cloud file as input data for the path planning system;

[0019] S12. Perform projection processing on the acquired 3D point cloud data, projecting the point cloud onto a 2D plane according to the height of the plane where the detection area is located, to obtain 2D projected point cloud data;

[0020] S13. Divide the two-dimensional projection area into sections based on the actual dimensions of the converter valve region. The grid cells are defined, and the grid is marked as occupied based on whether the projection point falls into the corresponding area;

[0021] S14. Calculate the equivalent expansion radius of the camera component based on the size of the camera component of the inspection robot, expand the occupied grid, and use the expanded grid map as the unified input data for the subsequent path accessibility detection module and local obstacle avoidance planning module.

[0022] Further, step S2 specifically includes:

[0023] S21. Determine the size of a single scanning area based on the horizontal and vertical scanning step lengths set according to the camera scanning window size, and ensure that there is a preset overlap area between adjacent scanning areas;

[0024] S22. Using a serpentine inspection method with alternating rows or a spiral inspection method, the system shrinks layer by layer from the outside to the inside to generate a global theoretical path with a serpentine or spiral scanning path.

[0025] Furthermore, the serpentine scanning path scans line by line in the row direction, starting from the beginning and moving one step vertically after scanning the end of a line. And then continue the inspection in the opposite direction on the next line, thus inspecting until the end.

[0026] Furthermore, the spiral scanning path starts from the outer boundary of the region and gradually contracts inward in a counterclockwise direction, finally converging at the center of the region, thus forming a progressive spiral structure for the inspection path.

[0027] Furthermore, step S3 specifically includes:

[0028] S31. For any two adjacent path points in the global theoretical path, use a parameterized line segment equation to represent them. Then, sample each path segment in the global theoretical path at a fixed interval. If the sampled point falls into the obstacle expansion region, the path segment is impassable.

[0029] S32. Perform continuous collision detection. If a path segment intersects linearly with a set of safety obstacles, then the path segment is impassable.

[0030] S33. If a path segment is impassable, set the last safe point in the corresponding path segment as the local starting point, and select the nearest path point that meets the safety requirements as the local target point along the global theoretical path.

[0031] Further, step S33 specifically includes: if the path segment (x) between any two adjacent path points... k y k ) to (x k+1 y k+1 If an obstacle is determined to be impassable, then the local obstacle avoidance starting point p is determined. s and local obstacle avoidance target point p g The starting point for local obstacle avoidance is taken from the path segment (x). k y k The last sampling point that is closest to the starting point and still belongs to the safety grid:

[0032] ;

[0033] Where n * Indicates the sequence number of the last safe sampling point in the path segment; Represents the nth path segment * The grid coordinates corresponding to each sampling point; This indicates that the sampling point did not enter the obstacle area; N s This represents the total number of samples; max{} indicates that the safe sampling point with the largest index is selected, which is the point that is closest to the obstacle but is still safe.

[0034] The local target point is the first passable point found along the theoretical path, either forward or backward from the end of a path segment. Let the local starting point be p. s =(i s j sLet the local target point be p. g =(i g j g ).

[0035] Further, step S4 specifically includes:

[0036] S41. Set the local search radius based on the expanded obstacle size, and construct a local search space in the area near the impassable path segment;

[0037] For the set V of all accessible raster nodes in the local search space, the total cost function for each node is:

[0038] f(n) = g(n) + h(n);

[0039] in g(n) represents the cumulative cost of bidirectional A* search at node n; h(n) is the dynamic heuristic function.

[0040] S42. Simultaneously initiate a bidirectional A* search from both the local starting point and the local target point;

[0041] S43. Determine the priority expansion direction based on the theoretical path direction vector, and only expand neighboring nodes whose direction deviation is less than a set threshold;

[0042] S44. Perform continuous collision detection on the lines connecting the candidate node and its parent node. If the line intersects with an obstacle, refuse to expand the node.

[0043] S45. Calculate the cost using a dynamic heuristic function that includes uncovered potential terms and security score terms.

[0044] Further, step S42 specifically includes:

[0045] A. First, set the local starting point p s =(i s j s As the starting node for the forward search, construct its open list OPEN. f With closed list CLOSED f ; Set the local target point p g =(i g j g As the starting node for the reverse search, construct its open list OPEN. b With closed list CLOSED b The initialization cost functions for the two search directions are defined as f(n) = g(n) + h(n), where g(n) = g f (n)+g b (n); the cumulative cost of the forward search is denoted as g.f (n), the cumulative cost of the backward search is denoted as g. b (n);

[0046] B. During the search process, A* searches in both directions are performed alternately. In each round of the search, starting from OPEN... f Select the node with the smallest cost function f(n) as the forward expansion node nf; then start from OPEN b The node with the minimum cost function is selected as the backward expansion node nb, and the neighborhoods of nf and nb are expanded respectively.

[0047] C. As the forward and backward searches continue to expand, when there exists any node n... * satisfy:

[0048] ,

[0049] Then, it is considered that the forward and backward searches meet, and the optimal meeting node n* is selected, satisfying...

[0050] ;

[0051] Where g'(n * ) represents node n * The corresponding minimum cost, g f (n) represents starting from the local starting point p. s The cumulative cost of forward search to node n; g b (n) represents the local target point p. g The cumulative cost of reverse search to node n;

[0052] D. Confirm the meeting node n * Then, backtrack from the meeting node along the parent node chain to the local starting point p. s The forward path sequence is obtained, and then backtracking is performed from the meeting node along the backward parent node chain to the local target point p. g Then, the path direction is reversed, and finally the forward sequence is concatenated with the reversed backward sequence.

[0053] Furthermore, the method is applied to an inspection robot, which adopts a T-shaped gantry two-axis right-angle mechanical structure and consists of a robot body 7, a camera assembly 1, an X-axis motion guide rail 6, and a Y-axis motion guide rail 8; obstacles and the camera assembly in the converter valve area are both located above the area of ​​the workpiece to be inspected.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] The inspection robot of this invention performs discrete sampling and continuous collision detection along each segment of the theoretical path to quickly identify impassable segments that conflict with expanding obstacles. It achieves a precise assessment of path safety, enabling the timely detection of potential collision risks before execution and providing accurate trigger conditions and start / end point positioning for local obstacle avoidance. Within a defined local area, the theoretical path direction vector is introduced as an extended constraint in the bidirectional search, ensuring that the direction of the local obstacle avoidance path remains consistent with the global inspection direction. This avoids significant deviations or reverse movements due to obstacle avoidance, ensuring the continuity and smoothness of the inspection task and reducing unnecessary robot posture adjustments and time consumption. The dynamic heuristic function incorporates coverage potential and safety scoring terms, enabling the search process to not only pursue the shortest path but also increase the scanning coverage area and move away from obstacles, avoiding getting trapped in local optima or deadlock. This achieves more intelligent and robust local path planning in complex, densely populated converter valve hall environments. Attached Figure Description

[0056] Figure 1 A flowchart of the global path planning method for a power converter valve inspection robot;

[0057] Figure 2 A 3D environment diagram of a power converter valve and a structural diagram of an inspection robot.

[0058] Figure 3 A two-dimensional grid map of a power converter valve containing obstacle information;

[0059] Figure 4 This is a schematic diagram of the operation path for a serpentine inspection route under a two-dimensional grid map.

[0060] Figure 5 This is a schematic diagram of the operation path for a spiral inspection route under a two-dimensional grid map;

[0061] Figure 6 Flowchart for path sampling and local region construction;

[0062] Figure 7 Flowchart of the improved A* bidirectional search local obstacle avoidance algorithm;

[0063] Figure 8 This is a schematic diagram of local obstacle avoidance path planning based on a dynamic heuristic function of coverage potential and safety score. Detailed Implementation

[0064] To more clearly describe the objectives, technical solutions, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0065] A global path planning method for a power converter valve inspection robot includes the following steps:

[0066] S1. Spatial data of the converter valve area is acquired using a 3D structured light camera, and the acquired 3D point cloud data is projected onto the plane of the detection area; a 2D raster map is constructed based on the projected data, and environmental obstacles are subjected to safety expansion processing to obtain a 2D raster map containing obstacle information for path planning. The point cloud data contains obstacles.

[0067] S2. Based on the horizontal and vertical scanning step lengths set according to the camera scanning window size, generate a serpentine or spiral global theoretical path to obtain a global theoretical path that ensures continuity and complete coverage of the region.

[0068] S3. Perform validity checks on path segments between adjacent path points along the global theoretical path, and use discrete sampling and continuous collision detection to determine whether the path segment overlaps with the expanded obstacles in the grid map; if the path segment is detected to be impassable, determine the starting point and target point of local obstacle avoidance.

[0069] S4. Construct a local search region centered on the region where the impassable path segment is located; simultaneously perform forward and backward A* search processes within the local search region; constrain the expandable neighborhood according to the theoretical path direction and perform continuous collision detection on candidate nodes; generate local obstacle avoidance paths using a dynamic heuristic function that includes coverage potential terms and safety score terms.

[0070] S5. Replace the corresponding path segment in the global theoretical path with the local obstacle avoidance path to form a new inspection path; the robot continues to perform subsequent inspection tasks along the updated global path until the entire area is inspected.

[0071] Optionally, S1: Spatial data of the converter valve area is acquired using a 3D structured light camera; the acquired 3D point cloud data is projected onto the plane of the detection area; a 2D raster map is constructed based on the projected data, and environmental obstacles are subjected to safety expansion processing to obtain a 2D raster map containing obstacle information for path planning, specifically including:

[0072] S11: Use a 3D structured light camera to acquire 3D point cloud data of obstacles in the converter valve area and store it as point cloud file as input data for the path planning system.

[0073] S12: Project the acquired 3D point cloud data, projecting the point cloud onto a 2D plane according to the height of the plane where the detection area is located, to obtain 2D projected point cloud data;

[0074] S13: Divide the two-dimensional projection area into sections based on the actual dimensions of the converter valve region. The grid cells are defined, and the grid is marked as occupied based on whether the projection point falls into the corresponding area;

[0075] S14: Calculate the equivalent expansion radius based on the size of the camera component of the inspection robot, expand the occupied grid, and use the expanded grid map as the unified input data for the subsequent path accessibility detection module and local obstacle avoidance planning module.

[0076] Optionally, S2: Based on the horizontal and vertical scanning step sizes set according to the camera scanning window size, a serpentine or spiral global theoretical path is generated to obtain a global theoretical path that ensures continuity and complete area coverage. Specifically, this includes:

[0077] S21: Determine the size of a single scanning area based on the horizontal and vertical scanning step lengths set according to the camera scanning window size, and ensure that there is a preset overlap area between adjacent scanning areas;

[0078] S22: Using a serpentine inspection method, the system alternates between rows or uses a spiral inspection method to shrink layer by layer from the outside to the inside, generating a global theoretical path with a serpentine scanning path or a spiral scanning path.

[0079] Optionally, S3 performs validity checks on path segments between adjacent path points along the global theoretical path, using discrete sampling and continuous collision detection to determine whether the path segment overlaps with expanded obstacles in the grid map; if a path segment is detected as impassable, the starting point and target point of local obstacle avoidance are determined, specifically including:

[0080] S31: For any two adjacent path points in the inspection path, use a parameterized line segment equation to represent them. Then, sample each path segment in the theoretical path at a fixed interval and determine whether the sampled points fall into the obstacle expansion area.

[0081] S32: Perform continuous collision detection to determine whether the path segment intersects linearly with the set of safety obstacles;

[0082] S33: When a path segment is impassable, set the last safe point in the path segment as the local starting point, and select the nearest path point that meets the safety requirements along the theoretical path as the local target point.

[0083] Optionally, S4 constructs a local search region centered on the region containing the impassable path segment; performs forward and backward A* search processes simultaneously within this region; constrains the expandable neighborhood based on the theoretical path direction and performs continuous collision detection on candidate nodes; and generates a local obstacle avoidance path using a dynamic heuristic function that includes a coverage potential term and a safety score term, specifically including:

[0084] S41: Set the local search radius based on the size of the expanded obstacle, and construct a local search space in the area near the impassable path segment;

[0085] S42: Simultaneously initiate a bidirectional A* search from both the local starting point and the local target point;

[0086] S43: Determine the priority expansion direction based on the theoretical path direction vector, and only expand neighboring nodes whose direction deviation is less than a set threshold.

[0087] S44: Perform continuous collision detection on the line connecting the candidate node and its parent node. If the line segment intersects with an obstacle, then refuse to expand the node.

[0088] S45: Calculate the cost using a dynamic heuristic function that includes uncovered potential terms and security score terms.

[0089] The overall flowchart of a global path planning method for a power converter valve inspection robot in this embodiment is as follows: Figure 1 As shown, a 3D structured light camera is first used to acquire 3D spatial data of the converter valve area, and the point cloud information is projected onto the horizontal plane of the detection area to construct a 2D grid map, providing a standardized environment for subsequent path planning. Based on this, a global theoretical path with continuity and coverage integrity is generated. Snake-like or spiral scanning methods can be used to ensure the inspection trajectory maintains the shortest path, improving inspection efficiency. During the robot's inspection along the theoretical path, a detection mechanism combining discrete sampling and continuous collision detection is used to determine the real-time validity of each path segment, accurately identifying impassable sections when encountering expanding obstacle areas. A local search range is constructed for the detected impassable path segments, and a local obstacle avoidance path is generated through bidirectional A* search. Finally, the local path after successful obstacle avoidance is stitched together with the global theoretical path to ensure the robot can safely and completely return to the preset theoretical path, thus forming a global inspection path with high coverage, high safety, and high feasibility.

[0090] The three-dimensional environment of the power converter valve and the structure of the inspection robot in this embodiment are as follows: Figure 2 As shown, the inspection robot adopts a T-shaped gantry two-axis right-angle mechanical structure, mainly composed of the robot body 7, camera assembly 1, X-axis motion guide rail 6, and Y-axis motion guide rail 8. Both axes use linear motion via lead screw transmission. Obstacles 2 and 4 within the converter valve area, along with the camera assembly, are located above the areas of the workpieces to be inspected 3 and 5. The collected 3D point cloud data of the obstacles is mapped onto a 2D reference plane at the same level as the inspection area using a vertical projection algorithm. A 2D grid map is constructed based on the projected obstacle information, and path planning is then performed on the 2D plane. This simplification is reasonable and efficient.

[0091] S11 utilizes a 3D structured light camera to acquire 3D point cloud data of obstacles within the converter valve area and stores it as a point cloud file as input data for the path planning system. Specifically:

[0092] A 3D structured light camera is used to scan the obstruction information of the converter valve, and a 3D point cloud set P consisting of K points is obtained:

[0093] ;

[0094] x t ,y t ,z t This represents the three-dimensional spatial coordinates of the t-th point in the point cloud relative to the camera coordinate system;

[0095] S12 performs projection processing on the acquired 3D point cloud data, projecting the point cloud onto a 2D plane according to the height of the plane containing the detection area, to obtain 2D projected point cloud data. Specifically:

[0096] Projecting all 3D point cloud data along its Z-axis (height direction) onto a horizontal 2D plane (Z=0) yields the projected point cloud P. 2D :

[0097] ;

[0098] S13 divides the two-dimensional projection area into M×N grid units according to the actual size of the converter valve area, and marks the grid units as occupied based on whether the projection point falls into the corresponding area. Specifically:

[0099] Two-dimensional grid map of power converter valve containing obstacle information, such as Figure 3 As shown. Based on the actual physical dimensions of the converter valve area, the inspection area is uniformly divided into an M×N discrete grid on a two-dimensional plane. The two-dimensional dimensions of the inspection area are set to 38×50 (unit: dm), i.e., M=38 (rows) and N=50 (columns). Each divided grid cell is considered as an independent grid, representing the smallest spatial unit in the environmental map. The grid set can be formally defined as:

[0100] ;

[0101] Where i represents the row index of the raster, and j represents the column index of the raster.

[0102] To facilitate data storage and path planning calculations, a column-first rule is used to map the aforementioned two-dimensional raster to a one-dimensional linear index L. The mapping relationship from two-dimensional coordinates (i,j) to the one-dimensional index L is as follows:

[0103] ;

[0104] Its inverse mapping relationship, that is, restoring the two-dimensional coordinates (i,j) based on the linear index L, is defined as:

[0105] ;

[0106] If a certain projection point cloud If a cell falls into a grid, the corresponding linear index is:

[0107] ;

[0108] Under the above linear numbering system, by collecting the linear indices corresponding to all occupied grid cells, we obtain the obstacle set:

[0109] ;

[0110] Where m represents the final number of obstacle grids formed by the projected point cloud; L k MN represents the linear number of the k-th obstacle grid cell; MN represents the total number of grid cells in the two-dimensional grid map.

[0111] S14 calculates the equivalent expansion radius of the camera component based on the size of the inspection robot's camera component, expands the occupied grid, and uses the expanded grid map as unified input data for the subsequent path accessibility detection module and local obstacle avoidance planning module. Specifically:

[0112] Based on the above inverse mapping relationship, the linear obstacle set L is transformed into a two-dimensional obstacle set Ω.

[0113] ;

[0114] Where m(i,j) is the occupancy function. If (i,j) is an obstacle, then m(i,j) = 1, otherwise it is 0.

[0115] Let the size of the camera on the two-dimensional plane be length L. p Width W p Then the equivalent radius of the camera can be approximately expressed as:

[0116] ;

[0117] Where r eff This represents the equivalent radius when the camera is projected onto a two-dimensional plane. The minimum safe distance is set to... Then the camera's overall safety radius is .

[0118] Based on the aforementioned safety radius, the original obstacle set Ω is expanded on a two-dimensional grid, resulting in the expanded safe obstacle set. Defined as:

[0119]

[0120] (i',j') represents the original obstacle grid coordinates; all grid cells that meet the conditions are considered as expanded obstacle cells;

[0121] The expanded grid is mapped to linear indices, resulting in the following representation of the linear index set of the expanded obstacles:

[0122]

[0123] Where Q is the index number in the set of expanding obstacles, and its value ranges from... ;

[0124] Represents a set The one-dimensional linear index value corresponding to the Qth inflated obstacle grid in the grid; Represents the set of expanding obstacles The number of grid cells; MN represents the total number of grid cells in a two-dimensional grid map;

[0125] S2 generates a serpentine or spiral global theoretical path based on the horizontal and vertical scanning step sizes set according to the camera scanning window size, thus obtaining a global theoretical path that ensures continuity and complete area coverage. Specifically:

[0126] Let the length of the camera scanning window be E, the width be F, and the corresponding movement step size be S. E S F The overlap rate between two adjacent scanned regions is 20%. For example... Figure 4 As shown, the serpentine inspection scans line by line in the row direction, starting from the beginning. When a line is scanned to the end, it moves one step S in the vertical direction. F And then continue the inspection in the opposite direction in the next row, thus forming a "round-trip" inspection until the end point. The path point sequence of the serpentine inspection can be represented as:

[0127] ;

[0128] Where (x0, y0) is the inspection starting point, located in the lower left of the area; (x N ,y1) is the inspection endpoint; N is the horizontal step number of a single-row inspection;

[0129] When the camera is in the Hth inspection zone, the generated lateral point sequence P t :

[0130]

[0131]

[0132] After completing one inspection zone, a line break is executed, and the above steps are repeated until the entire area is covered. Finally, a complete path point sequence P is generated along the serpentine inspection method.

[0133] like Figure 5 As shown, the spiral inspection starts from the outer boundary of the region and gradually contracts inward in a counterclockwise direction, finally converging at the center of the region, thus forming a progressively spiral structure for the inspection path. The path sequence points of the spiral inspection can be represented as:

[0134]

[0135] Where (x0, y0) is the inspection starting point, located in the lower left of the area; (x c ,y c ) represents the inspection endpoint; N represents the horizontal number of steps in a single-row inspection.

[0136] The spiral inspection starts at (x0, y0) and first inspects from left to right along the lower boundary. The horizontal sequence of points is as follows:

[0137]

[0138] Scanning from bottom to top along the right boundary, the vertical sequence points are:

[0139]

[0140] Where M is the number of longitudinal steps in a single-row inspection;

[0141] Scanning from right to left along the upper boundary, the horizontal sequence points are:

[0142]

[0143] Scanning from top to bottom along the left boundary, the vertical sequence points are:

[0144]

[0145] After completing one full scan, the boundary is reduced by one step size, and the above process is repeated until the spiral finally converges to the point (x) at the center of the region. c ,y c Finally, a complete path point sequence is generated along a spiral inspection method. .

[0146] S31 uses a parameterized line segment equation to represent any two adjacent path points in the path, then samples each path segment in the theoretical path at fixed intervals, and determines whether the sampled points fall into the obstacle expansion region. The flowchart for path sampling and local region construction is as follows: Figure 6 As shown. Specifically:

[0147] Let the global theoretical path be a sequence of raster points arranged sequentially:

[0148] ;

[0149] in The raster coordinates of the path points at time t are represented; T is the total number of points on the path.

[0150] For any adjacent path points in the path and The path segment, using linear parameterization, can be represented as:

[0151] ;

[0152] To perform discrete detection, according to sampling interval Discretize s as:

[0153] ;

[0154] The set of sampling points is obtained:

[0155] ;

[0156] Where △s represents the parameterized sampling step size, and the value of △s ranges from 0.1 to 0.25; This represents the grid coordinates corresponding to the nth sampling point on the path segment;

[0157] Then, an occupancy determination is performed on each sampling point, if its corresponding raster linear index is:

[0158] ;

[0159] If satisfied If the condition is met, the path segment is deemed impassable; otherwise, it is permissible.

[0160] S32 performs continuous collision detection to determine whether a path segment linearly intersects with a set of safety obstacles. Specifically:

[0161] When discrete sampling detects a path as passable or close to the boundary, continuous collision detection is further employed to avoid missed detections. The path segment is parameterized as a vector in the continuous grid coordinate system as follows:

[0162]

[0163] Coordinates of each original obstacle grid Calculate the shortest distance:

[0164]

[0165] If there exists any (i', j') such that If the path segment intersects (collides) with an obstacle continuously, then the segment is deemed impassable.

[0166] When a path segment is impassable, step S33 sets the last safe point in the path segment as the local starting point and selects the nearest path point that meets the safety requirements along the theoretical path as the local target point. Specifically:

[0167] If the path segment (x) between any two adjacent path points k y k ) to (x k+1 y k+1 If an obstacle is determined to be impassable, then the local obstacle avoidance starting point p is determined. s and local obstacle avoidance target point p g The starting point for local obstacle avoidance is taken from the path segment (x). k y k The last sampling point that is closest to the starting point and still belongs to the safety grid:

[0168]

[0169] Where n * Indicates the sequence number of the last safe sampling point in the path segment; Represents the nth path segment * The grid coordinates corresponding to each sampling point; This indicates that the sampling point did not enter the obstacle area; N s This represents the total number of samples; max{} indicates that the safe sampling point with the largest index is selected, which is the point that is closest to the obstacle but is still safe.

[0170] The local target point is the first passable point found along the theoretical path, either forward or backward from the end of a path segment. Let the local starting point be p. s =(i s j s Let the local target point be p. g =(i g j g ).

[0171] S41 sets the local search radius based on the expanded obstacle size, and constructs a local search space in the area near the impassable path segment. Specifically:

[0172] by Construct a structure with radius R centered at the origin. local The local search region (circular or square) is defined as:

[0173] ;

[0174] in K is an empirical coefficient (usually taken as 3~6); (i s j s ) is p s Grid coordinates;

[0175] In the aforementioned local areas Within this framework, obstacle avoidance planning is modeled as an optimal path search problem based on a grid graph. A local grid graph is constructed as follows:

[0176] ;

[0177] in , represents the set of all locally accessible grid nodes; E indicates that the node connections are constructed using the 8-neighborhood rule to enhance path smoothness;

[0178] The total cost function for each node in V is:

[0179] f(n) = g(n) + h(n);

[0180] Where g(n) represents the cumulative cost of bidirectional A* search at node n; h(n) is the dynamic heuristic function;

[0181] S42 initiates a bidirectional A* search simultaneously from both the local starting point and the local target point. Specifically:

[0182] After constructing the local search region, a bidirectional A* search is used within this region for local obstacle avoidance path planning. The flowchart of the improved A* bidirectional search local obstacle avoidance algorithm is as follows: Figure 7 As shown. Includes the following steps:

[0183] A. First, set the local starting point p s =(i s j s As the starting node for the forward search, construct its open list OPEN. f With closed list CLOSED f ; Set the local target point p g =(i g j g As the starting node for the reverse search, construct its open list OPEN. b With closed list CLOSED b The initialization cost functions for the two search directions are defined as f(n) = g(n) + h(n), where g(n) = g f (n)+g b (n); the cumulative cost of the forward search is denoted as g. f (n), the cumulative cost of the backward search is denoted as g. b (n);

[0184] B. During the search process, A* searches in both directions are performed alternately. In each round of the search, starting from OPEN... f Select the node with the smallest cost function f(n) as the forward expansion node nf; then start from OPEN b The node with the minimum cost function is selected as the backward expansion node nb, and the neighborhoods of nf and nb are expanded respectively.

[0185] C. As the forward and backward searches continue to expand, when there exists any node n... * satisfy:

[0186]

[0187] Then, it is assumed that the forward and backward searches meet, and the optimal meeting node n is selected. * ,satisfy

[0188]

[0189] Where g'(n*) represents the minimum cost corresponding to node n*, g f (n) represents starting from the local starting point p. s The cumulative cost of forward search to node n; g b (n) represents the local target point p. g The cumulative cost of reverse search to node n;

[0190] D. Confirm the meeting node n * Then, backtrack from the meeting node along the parent node chain to the local starting point p. s The forward path sequence is obtained, and then backtracking is performed from the meeting node along the backward parent node chain to the local target point p. g Then, the path direction is reversed, and finally the forward sequence is concatenated with the reversed backward sequence.

[0191] S43 determines the preferred expansion direction based on the theoretical path direction vector, and only expands neighboring nodes whose direction deviation is less than a set threshold. Specifically, let the direction vector of the theoretical path at the local starting point be:

[0192]

[0193] Let the coordinates of the parent node be... For any candidate neighbor node n, its movement direction vector is:

[0194]

[0195] The directional deviation angle is:

[0196]

[0197] like If the condition is met, then the expansion of the neighboring node is allowed; otherwise, expansion is prohibited.

[0198] Where φ is the direction threshold angle, which ranges from 60° to 90°.

[0199] S44 performs continuous collision detection on the lines connecting candidate nodes and their parent nodes. If a line segment intersects with an obstacle, the node is rejected from expansion. Specifically:

[0200] Let the parameterized expression of the line segment be:

[0201]

[0202] Discrete sampling is performed on t. If there exists t=t * Make:

[0203]

[0204] Then the node expansion will be rejected;

[0205] in This represents the expanded set of obstacles; Indicates the parent node; n is the candidate node;

[0206] S45 uses a dynamic heuristic function that includes uncovered potential terms and security scoring terms to calculate the cost. Specifically:

[0207] Let the expression for the dynamic heuristic function be:

[0208] h(n) = βU(n) - γS(n)

[0209] Where β represents the weighting coefficient of the coverage potential item, β≥0; γ represents the weighting coefficient of the safety scoring item, γ≥0;

[0210] The coverage potential term U(n) measures whether there are unscanned areas around a node, encouraging local obstacle avoidance paths to traverse uncovered areas as much as possible while meeting safety requirements. It is defined as follows:

[0211]

[0212] Where N R (n) represents the set of neighboring grid cells centered at node n with radius R; c(i,j)=1 indicates that grid cell (i,j) has been covered by the theoretical path, and c(i,j)=0 indicates that grid cell (i,j) has not been covered; R is the search radius of the covered neighborhood, which is usually 2 to 4 grid cells.

[0213] The safety score term reflects the relative safety of a node relative to the expanded set of obstacles, in order to prevent local paths from getting too close to obstacles. The safety score term is defined as follows.

[0214]

[0215] Where (i n ,j n (i',j') represents the grid coordinates of the node; (i',j') represents the set of dilatational obstacles. Any obstacle grid in the grid;

[0216] The aforementioned dynamic heuristic function comprehensively constrains and guides the node expansion order during the local obstacle avoidance search process by simultaneously introducing a coverage potential term and a safety score term. The coverage potential term U(n) reflects the number of unscanned grids in the neighborhood surrounding a node. A higher value indicates a greater contribution of the node's area to subsequent inspection tasks. By introducing this term into the heuristic function, nodes that help fill uncovered areas can be prioritized during local obstacle avoidance, thus avoiding the problems of repeated traversal and insufficient coverage caused by traditional obstacle avoidance methods that only focus on bypassing obstacles. This ensures that the obstacle avoidance path satisfies both feasibility and inspection coverage integrity. The safety score term S(n) measures the minimum distance between a node and the expanding obstacle set, reflecting the safety margin of the node's location. By applying a penalty weight to the safety score term in the heuristic function, excessive proximity to obstacle edges during the search process can be effectively suppressed, reducing search failures due to insufficient safety distances, thereby improving the stability and success rate of local path planning.

[0217] like Figure 8 As shown, the local obstacle avoidance path planning process based on bidirectional A* search and incorporating a dynamic heuristic function is presented within a local search region. The figure uses a 2D grid map as a basis, where obstacles are expanded to form impassable regions, surrounded by passable grid regions. Within this local search region, a local starting point is set. With local target point These serve as the starting nodes for the forward and backward searches, respectively. The forward search starts from the local starting point and gradually expands towards the target; the backward search starts from the local target point and expands back towards the starting point. The candidate nodes are marked in the diagram. , , This represents the positions of different neighboring nodes within the same search round. The nodes... Located near an uncovered area, its neighborhood contains many rasters that have not yet been scanned, indicating high coverage potential; Node Located near the boundary of the expanding obstacle, and relatively close to the obstacle; node Positioned between these two extremes, this node maintains a safe distance from obstacles while still containing a certain number of uncovered grid cells in its neighborhood, balancing coverage potential and safety. During the search process, a dynamic heuristic function value, including a coverage potential term and a safety score term, is calculated for each candidate node. The coverage potential term measures the number of uncovered areas in the node's neighborhood, guiding the search to prioritize node directions that contribute to subsequent inspection coverage. The safety score term reflects the distance relationship between the node and the expanding obstacle, negatively impacting nodes that are too close to the obstacle. Through a comprehensive weighing of these two factors, the algorithm tends to select nodes... As a priority expansion node.

[0218] In the heuristic function, the coverage potential term U(n) and the safety score term S(n) are not independent but complementary: the coverage potential term focuses on guiding the search results to maximize the long-term benefits of the global inspection task, while the safety score term focuses on constraining the feasibility and safety of the search process. The dynamic heuristic function used in this invention, while ensuring a reasonable safe distance from obstacles, guides the search to prioritize node directions that contribute to subsequent inspection coverage, thereby reducing the number of invalid search nodes and achieving a simultaneous improvement in obstacle avoidance efficiency and inspection coverage.

[0219] S5 replaces the corresponding path segment in the global theoretical path with the local obstacle avoidance path, forming a new inspection path; the robot continues to perform subsequent inspection tasks along the updated global path until the entire area is inspected. Specifically:

[0220] First, determine the starting point and target point of the local obstacle avoidance in the theoretical path; then, using the local obstacle avoidance path as a replacement segment, delete the original path segment between the two in the theoretical path, and insert the new path after obstacle avoidance at that position to form a continuous, uninterrupted global feasible path.

[0221] After acquiring a new globally feasible path, the robot will continue its inspection task along the updated path. Once the robot completes the inspection coverage of the current area along the updated path, the system automatically enters the feasibility detection process for the next path segment, repeating steps S3 to S5 until the entire converter valve area is completely scanned and covered.

[0222] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A global path planning method for a power converter valve inspection robot, comprising the following steps: S1. Use a 3D structured light camera to acquire 3D point cloud data of the converter valve area, and project the acquired 3D point cloud data onto the plane where the detection area is located; construct a 2D raster map based on the projected data, and perform safety expansion processing on environmental obstacles to obtain a 2D raster map containing obstacle information and used for path planning. S2. Generate a serpentine or spiral global theoretical path based on the horizontal and vertical scanning step sizes set according to the camera scanning window size; S3. Perform validity checks on the path segments between adjacent path points along the global theoretical path, and use discrete sampling and continuous collision detection to determine whether the path segments overlap with the expanded obstacles in the two-dimensional grid map; if the path segment is detected to be impassable, determine the starting point and target point of local obstacle avoidance; S4. Construct a local search region centered on the region where the impassable path segment is located; simultaneously perform forward and backward A* search processes within the local search region; constrain the expandable neighborhood based on the global theoretical path direction, and perform continuous collision detection on candidate nodes; generate local obstacle avoidance paths using a dynamic heuristic function that includes coverage potential and safety score terms; The expression for the dynamic heuristic function is: h(n) = βU(n) - γS(n); Where β represents the weighting coefficient of the coverage potential term; γ represents the weighting coefficient of the safety score term; U(n) represents the coverage potential term, which measures whether there are areas around the node that have not yet been scanned; S(n) represents the safety score term, which reflects the relative safety of the node and the set of expanding obstacles. S5. Replace the corresponding path segment in the global theoretical path with the local obstacle avoidance path to form a new global theoretical path; the robot performs subsequent inspection tasks along the updated global theoretical path until the entire area is covered by inspection.

2. The global path planning method for a power converter valve inspection robot according to claim 1, characterized in that, Step S1 specifically includes: S11. Use a 3D structured light camera to acquire 3D point cloud data of obstacles in the converter valve area, and store it as point cloud file as input data for the path planning system; S12. Perform projection processing on the acquired 3D point cloud data, projecting the point cloud onto a 2D plane according to the height of the plane where the detection area is located, to obtain 2D projected point cloud data; S13. Divide the two-dimensional projection area into sections based on the actual dimensions of the converter valve region. The grid cells are defined, and the grid is marked as occupied based on whether the projection point falls into the corresponding area; S14. Calculate the equivalent expansion radius of the camera component based on the size of the camera component of the inspection robot, expand the occupied grid, and use the expanded grid map as the unified input data for the subsequent path accessibility detection module and local obstacle avoidance planning module.

3. The global path planning method for a power converter valve inspection robot according to claim 1, characterized in that, Step S2 specifically includes: S21. Determine the size of a single scanning area based on the horizontal and vertical scanning step lengths set according to the camera scanning window size, and ensure that there is a preset overlap area between adjacent scanning areas; S22. Using a serpentine inspection method with alternating rows or a spiral inspection method, the system shrinks layer by layer from the outside to the inside to generate a global theoretical path with a serpentine or spiral scanning path.

4. The global path planning method for a power converter valve inspection robot according to claim 3, characterized in that, The serpentine scanning path scans line by line in the row direction, starting from the beginning. When a line is scanned to the end, it moves one step in the vertical direction. And then continue the inspection in the opposite direction on the next line, thus inspecting until the end.

5. The global path planning method for a power converter valve inspection robot according to claim 3, characterized in that, The spiral scanning path starts from the outer boundary of the region and gradually contracts inward in a counterclockwise direction, finally converging at the center of the region, thus forming a progressive spiral structure for the inspection path.

6. The global path planning method for a power converter valve inspection robot according to claim 1, characterized in that, Step S3 specifically includes: S31. For any two adjacent path points in the global theoretical path, use a parameterized line segment equation to represent them. Then, sample each path segment in the global theoretical path at a fixed interval. If the sampled point falls into the obstacle expansion region, the path segment is impassable. S32. Perform continuous collision detection. If a path segment intersects linearly with a set of safety obstacles, then the path segment is impassable. S33. If a path segment is impassable, set the last safe point in the corresponding path segment as the local starting point, and select the nearest path point that meets the safety requirements as the local target point along the global theoretical path.

7. The global path planning method for a power converter valve inspection robot according to claim 6, characterized in that, Step S33 specifically includes: if the path segment (x) between any two adjacent path points k y k ) to (x k+1 y k+1 If an obstacle is determined to be impassable, then the local obstacle avoidance starting point p is determined. s and local obstacle avoidance target point p g The starting point for local obstacle avoidance is taken from the path segment (x). k y k The last sampling point that is closest to the starting point and still belongs to the safety grid: ； Where n * Indicates the sequence number of the last safe sampling point in the path segment; Represents the nth path segment * The grid coordinates corresponding to each sampling point; This indicates that the sampling point did not enter the obstacle area; N s This represents the total number of samples; max{} indicates that the safe sampling point with the largest index is selected, which is the point that is closest to the obstacle but is still safe. The local target point is the first passable point found along the theoretical path, either forward or backward from the end of a path segment. Let the local starting point be p. s =(i s j s Let the local target point be p. g =(i g j g ).

8. The global path planning method for a power converter valve inspection robot according to claim 1, characterized in that, Step S4 specifically includes: S41. Set the local search radius based on the expanded obstacle size, and construct a local search space in the area near the impassable path segment; For the set V of all accessible raster nodes in the local search space, the total cost function for each node is: f(n) = g(n) + h(n); in Let (i, j) be the coordinates of all accessible raster nodes in the local search space, and R be the coordinates of the raster nodes. local For the local search radius, Let g(n) be the expanded set of safety barriers; g(n) represents the cumulative cost of bidirectional A* search at node n; h(n) is the dynamic heuristic function. S42. Simultaneously initiate a bidirectional A* search from both the local starting point and the local target point; S43. Determine the priority expansion direction based on the theoretical path direction vector, and only expand neighboring nodes whose direction deviation is less than a set threshold; S44. Perform continuous collision detection on the lines connecting the candidate node and its parent node. If the line intersects with an obstacle, refuse to expand the node. S45. Calculate the cost using a dynamic heuristic function that includes uncovered potential terms and security score terms.

9. The global path planning method for a power converter valve inspection robot according to claim 8, characterized in that, Step S42 specifically includes: A. First, set the local starting point p s =(i s j s As the starting node for the forward search, construct its open list OPEN. f With closed list CLOSED f ; Set the local target point p g =(i g j g As the starting node for the reverse search, construct its open list OPEN. b With closed list CLOSED b The initialization cost functions for the two search directions are defined as f(n) = g(n) + h(n), where g(n) = g f (n)+g b (n); the cumulative cost of the forward search is denoted as g. f (n), the cumulative cost of the backward search is denoted as g. b (n); B. During the search process, A* searches in both directions are performed alternately. In each round of the search, starting from OPEN... f Select the node with the smallest cost function f(n) as the forward expansion node nf; then start from OPEN b The node with the minimum cost function is selected as the backward expansion node nb, and the neighborhoods of nf and nb are expanded respectively. C. As the forward and backward searches continue to expand, when there exists any node n... * satisfy: ； Then, it is assumed that the forward and backward searches meet, and the optimal meeting node n is selected. * ,satisfy: ； Where g'(n * ) represents node n * The corresponding minimum cost, g f (n) represents starting from the local starting point p. s The cumulative cost of forward search to node n; g b (n) represents the local target point p. g The cumulative cost of reverse search to node n; D. Confirm the meeting node n * Then, backtrack from the meeting node along the parent node chain to the local starting point p. s The forward path sequence is obtained, and then backtracking is performed from the meeting node along the backward parent node chain to the local target point p. g Then, the path direction is reversed, and finally the forward sequence is concatenated with the reversed backward sequence.

10. An inspection robot employing the global path planning method as described in any one of claims 1-9, characterized in that, The inspection robot adopts a T-shaped gantry two-axis right-angle mechanical structure, consisting of the robot body (7), camera assembly (1), X-axis motion guide rail (6) and Y-axis motion guide rail (8); obstacles and camera assembly in the converter valve area are all located above the workpiece area to be inspected.

Images