A path planning method of a multi-axis linkage electric arc cladding robot

Abstract

The application provides a path planning method of a multi-axis linkage electric arc cladding robot, comprising collecting point cloud data of a workpiece to be cladded and building an octree structure, completing initial region pruning according to leaf node point cloud density and normal vector dispersion, transmitting the normal vector dispersion to represent the complexity of a surface, adjusting the pruning threshold value as the weight, completing point cloud simplification from top to bottom, extracting candidate key points, solving the inverse kinematic solution of a welding gun pose, constructing a motion cost function conforming to joint limiting, screening an optimal pose following inverse solution branch consistency, eliminating abnormal points to form a rough screening key point set, building a normalized comprehensive evaluation function integrating welding gun pose smoothness, node spacing uniformity, local heat input uniformity and motion cost, and iteratively optimizing under the constraint of geometric fidelity until the function converges to determine a key point sequence, and generating a stable electric arc cladding motion path through Cartesian space interpolation.

Description

Technical Field

[0001] This application belongs to the field of robotics, and in particular relates to a path planning method for a multi-axis linkage arc cladding robot. Background Technology

[0002] To adapt to the increasing complexity and diversity of surface morphology of workpieces to be clad, multi-axis robots equipped with welding torches are typically used for continuous spatial operations in actual processing. For metal workpieces with free-form surfaces and complex local features, planning a path that balances processing efficiency, robot motion constraints, and cladding quality is a crucial issue in robot cladding path planning. Existing path planning processes usually require acquiring 3D point cloud data of the workpiece surface first. However, the original point cloud is large in scale, and directly using it in path calculation would significantly increase the computational burden and prolong the planning cycle. Using conventional proportional downsampling methods may weaken the geometric representation of high-curvature regions, edge regions, and regions with local abrupt changes, affecting the path's accuracy in fitting complex surfaces. Furthermore, the motion space of multi-axis robots exhibits strong nonlinearity. If the trajectory is generated solely based on the geometric model in the early stages of path planning without fully considering joint mechanism limitations, inverse motion multi-branching, and the continuity of adjacent path point configurations, problems such as abrupt changes in welding torch posture, discontinuous joint motion, approaching mechanical limits, or even path unexecutability may occur during actual processing, affecting the stability and safety of the cladding process.

[0003] Traditional point cloud pruning strategies often rely on a globally fixed single density threshold, making it difficult to fully integrate local surface complexity for hierarchical simplification. This often results in the retention of redundant point clouds in flat areas while insufficient preservation of key geometric features in complex curvature regions. Existing trajectory key point extraction methods typically prioritize geometric fidelity, with robot motion inverse kinematics often lagging behind trajectory generation. This leads to a lack of effective coordination between point cloud spatial distribution, path geometry, and robot motion costs, making it difficult to identify potential risks such as inconsistencies in inverse kinematics branches and joint approach limits during the key point selection stage. Furthermore, existing path generation methods have limited ability to jointly optimize multi-dimensional indicators, often failing to incorporate process and equipment constraints such as welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost into the evaluation function for iterative optimization. While the resulting Cartesian space path may meet basic surface coverage requirements, there is still room for improvement in heat input stability, robot motion smoothness, and complex surface fit, potentially affecting the forming quality and processing consistency of the cladding layer. Summary of the Invention

[0004] To address the problems in existing technologies, such as the loss of complex surface features in point cloud refinement, insufficient consideration of robot motion constraints in path planning, and inadequate thermal input stability, this invention provides a path planning method and system for a multi-axis linkage arc cladding robot.

[0005] In a first aspect, the present invention provides the following technical solution: a path planning method for a multi-axis linkage arc cladding robot, comprising the following steps: Obtain point cloud data of the workpiece to be clad and construct an octree structure; calculate the point cloud density and normal vector dispersion within the voxels of the leaf nodes of the octree, and perform initial pruning on voxels whose point cloud density is lower than a first threshold and whose normal vector dispersion is lower than a second threshold as non-critical regions; pass the normal vector dispersion of the child nodes up to the parent node to represent the surface complexity; use the surface complexity of the parent node to adjust the pruning threshold of the child nodes to perform top-down pruning to obtain a simplified point cloud; Based on the simplified point cloud, candidate key points are extracted, the inverse motion solution of the welding gun pose corresponding to each candidate key point is calculated, and a motion cost function inversely proportional to the joint rotation angle distance from the mechanism limit distance is constructed. Under the premise of satisfying the consistency of the inverse solution branches, the solution with the lowest cost function value is selected as the optimal solution, and candidate key points with motion cost or geometric deviation exceeding the threshold are eliminated to obtain a coarse key point set. A comprehensive performance evaluation function is constructed, which includes the normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost. Using the comprehensive performance evaluation function as the objective, under geometric fidelity constraints, the coarse-screened key point set is iteratively eliminated and optimized until the comprehensive performance evaluation function value converges, resulting in a key point sequence. The key point sequence is then interpolated in Cartesian space to generate the cladding path.

[0006] Optionally, acquiring the point cloud data of the workpiece to be clad and constructing an octree structure includes: Scan the surface of the workpiece to be clad to obtain the original three-dimensional point cloud data; Calculate the extreme values ​​of the original 3D point cloud data in the three coordinate axes, generate an initial bounding box that surrounds the original 3D point cloud data, and use the initial bounding box as the root node of the octree; The space corresponding to the root node is divided in half along the midpoint of the three coordinate axes to generate eight subspaces of the same volume as child nodes. The number of point clouds contained in each child node is recursively determined. If the number of point clouds in the child node is greater than a preset threshold and the spatial size of the child node is greater than a preset lower limit, the child node is further divided into equal parts until all leaf nodes satisfy the condition that the number of point clouds is not greater than the threshold or the spatial size reaches the lower limit.

[0007] Optionally, the step of calculating the point cloud density and normal vector dispersion within the voxels of the leaf nodes of the octree includes: Traverse the bottom-level leaf nodes of the octree and count the total number of discrete point clouds contained in each leaf node voxel. The point cloud density of the voxel is obtained by dividing the total number of point clouds within the leaf node voxel by the volume of the voxel. Principal component analysis is used to calculate the local normal vectors of each point cloud within the leaf node voxel. The local normal vectors of all point clouds within the voxel are then uniformly oriented and normalized. The variance of the angle between each local normal vector and the average normal vector within the voxel is calculated, and this variance is used as the normal vector dispersion of the voxel.

[0008] Optionally, the step of solving the inverse motion solution of the welding torch pose corresponding to each candidate key point and constructing a motion cost function inversely proportional to the joint rotation angle distance from the mechanism limit distance includes: Based on the robot's link parameters and motion constraints, multiple sets of selectable joint angle solutions for the target pose corresponding to the current candidate keypoint are obtained by solving the inverse motion matrix. For each joint angle solution, the current calculated angle of each joint of the robot is extracted, the absolute value of the difference between the current calculated angle and the upper and lower limit boundaries of the joint for safe operation is calculated, and the smallest absolute value of the difference is selected as the limit margin of the joint. The weighted sum of the limit margin of each joint and the reciprocal of the preset positive bias factor is used as the motion cost function value of the joint angle solution.

[0009] Optionally, the construction of the comprehensive performance evaluation function includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost, including: According to the expression Calculate the comprehensive performance evaluation function value; Where E represents the value of the comprehensive performance evaluation function; S, D, H, and K represent the welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost after normalization, respectively. Among them, the smoothness of the welding torch posture is calculated by the standard deviation of the rate of change of the welding torch posture angle between adjacent key points; The uniformity of node spacing is calculated using the standard deviation of the distance between adjacent key points; The uniformity of local heat input is calculated by the dispersion coefficient of the heat input per unit length corresponding to each adjacent key point path segment. The heat input per unit length is determined based on the cladding current, cladding voltage, welding torch moving speed and thermal efficiency coefficient. The average motion cost is calculated by averaging the motion costs of the optimal solutions for all key points in the path sequence; These are the preset weighting coefficients for the four evaluation components.

[0010] Optionally, the step of iteratively eliminating and optimizing the set of key points for coarse screening under geometric fidelity constraints, with the comprehensive performance evaluation function as the objective, until the comprehensive performance evaluation function converges, to obtain the key point sequence, includes: Set the maximum number of iterations; In each iteration, we attempt to temporarily remove a single key point from the current set of coarse screening key points, and recalculate the comprehensive performance evaluation function value after removing the key point. Comparing all temporary removal schemes, under the constraint that the geometric deviation between the path generated after removing the key points and the initial point cloud model is less than the set tolerance, the key points that cause the largest decrease in the comprehensive performance evaluation function value are selected for permanent removal, and the current key point set is updated. Repeat the above steps of temporary removal and permanent elimination until the decrease in the comprehensive performance evaluation function value is less than the preset convergence threshold or the maximum number of iterations is reached, and then take the key point set at this time as the key point sequence.

[0011] Optionally, the step of performing Cartesian space interpolation on the key point sequence to generate the cladding path includes: Extract the Cartesian space pose of each point in the keypoint sequence; Between adjacent key points, a Cartesian space interpolation algorithm is used to smooth the transition of the Cartesian space pose sequence, generating a Cartesian control trajectory with smooth transition in both position and attitude. Combining the optimal angle sequence of the robot joint space corresponding to each point in the key point sequence, the optimal joint space solution of adjacent key points is used as the initial value or branch constraint for inverse motion solution. The Cartesian control trajectory is mapped to the joint space through the robot inverse motion model. The mapped joint trajectory is then checked for joint limit, inverse solution branch consistency, joint velocity and joint acceleration. A joint control trajectory that meets the preset motion constraints is generated, and a smooth and continuous cladding path is output.

[0012] Secondly, the present invention also provides the following technical solution: a path planning system for a multi-axis linkage arc cladding robot, comprising the following modules: The execution module is used to acquire point cloud data of the workpiece to be clad and construct an octree structure; calculate the point cloud density and normal vector dispersion within the voxels in the leaf nodes of the octree, and perform initial pruning on voxels whose point cloud density is lower than a first threshold and whose normal vector dispersion is lower than a second threshold as non-critical regions; pass the normal vector dispersion of the child nodes up to the parent node to represent the surface complexity; and perform top-down pruning by adjusting the pruning threshold of the child nodes using the surface complexity of the parent node to obtain a simplified point cloud. The extraction module is used to extract candidate key points based on the simplified point cloud, calculate the inverse motion solution of the welding torch pose corresponding to each candidate key point, and construct a motion cost function that is inversely proportional to the joint rotation angle distance from the mechanism limit distance. Under the premise of satisfying the consistency of the inverse solution branches, the solution with the lowest cost function value is selected as the optimal solution, and candidate key points with motion cost or geometric deviation exceeding the threshold are eliminated to obtain a coarse key point set. A generation module is used to construct a comprehensive performance evaluation function, which includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost. Using the comprehensive performance evaluation function as the objective, under geometric fidelity constraints, the coarse-screened key point set is iteratively eliminated and optimized until the comprehensive performance evaluation function value converges, resulting in a key point sequence. The key point sequence is then interpolated in Cartesian space to generate the cladding path.

[0013] Preferably, the step of acquiring point cloud data of the workpiece to be clad and constructing an octree structure includes: Scan the surface of the workpiece to be clad to obtain the original three-dimensional point cloud data; Calculate the extreme values ​​of the original 3D point cloud data in the three coordinate axes, generate an initial bounding box that surrounds the original 3D point cloud data, and use the initial bounding box as the root node of the octree; The space corresponding to the root node is divided in half along the midpoint of the three coordinate axes to generate eight subspaces of the same volume as child nodes. The number of point clouds contained in each child node is recursively determined. If the number of point clouds in the child node is greater than a preset threshold and the spatial size of the child node is greater than a preset lower limit, the child node is further divided into equal parts until all leaf nodes satisfy the condition that the number of point clouds is not greater than the threshold or the spatial size reaches the lower limit.

[0014] Preferably, the step of calculating the point cloud density and normal vector dispersion within the voxels of the leaf nodes of the octree includes: Traverse the bottom-level leaf nodes of the octree and count the total number of discrete point clouds contained in each leaf node voxel. The point cloud density of the voxel is obtained by dividing the total number of point clouds within the leaf node voxel by the volume of the voxel. Principal component analysis is used to calculate the local normal vectors of each point cloud within the leaf node voxel. The local normal vectors of all point clouds within the voxel are then uniformly oriented and normalized. The variance of the angle between each local normal vector and the average normal vector within the voxel is calculated, and this variance is used as the normal vector dispersion of the voxel.

[0015] Preferably, the step of solving the inverse motion solution of the welding torch pose corresponding to each candidate key point and constructing a motion cost function inversely proportional to the joint rotation distance from the mechanism limit distance includes: Based on the robot's link parameters and motion constraints, multiple sets of selectable joint angle solutions for the target pose corresponding to the current candidate keypoint are obtained by solving the inverse motion matrix. For each joint angle solution, the current calculated angle of each joint of the robot is extracted, the absolute value of the difference between the current calculated angle and the upper and lower limit boundaries of the joint for safe operation is calculated, and the smallest absolute value of the difference is selected as the limit margin of the joint. The weighted sum of the limit margin of each joint and the reciprocal of the preset positive bias factor is used as the motion cost function value of the joint angle solution.

[0016] Preferably, the construction of the comprehensive performance evaluation function includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost, including: According to the expression Calculate the comprehensive performance evaluation function value; Where E represents the value of the comprehensive performance evaluation function; S, D, H, and K represent the welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost after normalization, respectively. Among them, the smoothness of the welding torch posture is calculated by the standard deviation of the rate of change of the welding torch posture angle between adjacent key points; The uniformity of node spacing is calculated using the standard deviation of the distance between adjacent key points; The uniformity of local heat input is calculated by the dispersion coefficient of the heat input per unit length corresponding to each adjacent key point path segment. The heat input per unit length is determined based on the cladding current, cladding voltage, welding torch moving speed and thermal efficiency coefficient. The average motion cost is calculated by averaging the motion costs of the optimal solutions for all key points in the path sequence; These are the preset weighting coefficients for the four evaluation components.

[0017] Preferably, the step of iteratively eliminating and optimizing the set of key points for coarse screening under geometrical fidelity constraints, with the comprehensive performance evaluation function as the objective, until the comprehensive performance evaluation function converges, to obtain the key point sequence, includes: Set the maximum number of iterations; In each iteration, we attempt to temporarily remove a single key point from the current set of coarse screening key points, and recalculate the comprehensive performance evaluation function value after removing the key point. Comparing all temporary removal schemes, under the constraint that the geometric deviation between the path generated after removing the key points and the initial point cloud model is less than the set tolerance, the key points that cause the largest decrease in the comprehensive performance evaluation function value are selected for permanent removal, and the current key point set is updated. Repeat the above steps of temporary removal and permanent elimination until the decrease in the comprehensive performance evaluation function value is less than the preset convergence threshold or the maximum number of iterations is reached, and then take the key point set at this time as the key point sequence.

[0018] Preferably, the step of performing Cartesian space interpolation on the key point sequence to generate the cladding path includes: Extract the Cartesian space pose of each point in the keypoint sequence; Between adjacent key points, a Cartesian space interpolation algorithm is used to smooth the transition of the Cartesian space pose sequence, generating a Cartesian control trajectory with smooth transition in both position and attitude. Combining the optimal angle sequence of the robot joint space corresponding to each point in the key point sequence, the optimal joint space solution of adjacent key points is used as the initial value or branch constraint for inverse motion solution. The Cartesian control trajectory is mapped to the joint space through the robot inverse motion model. The mapped joint trajectory is then checked for joint limit, inverse solution branch consistency, joint velocity and joint acceleration. A joint control trajectory that meets the preset motion constraints is generated, and a smooth and continuous cladding path is output.

[0019] This invention constructs an octree structure and performs point cloud pruning by combining point cloud density and normal vector dispersion. Simultaneously, it dynamically adjusts the pruning threshold using surface complexity, reducing redundant point cloud data while preserving the geometric features of complex surface regions on the workpiece, thus improving the data processing efficiency of subsequent path planning. By introducing a motion cost function based on the joint mechanism's constraint margin and selecting a superior inverse motion solution under the constraint of inverse solution branch consistency, the risk of joints approaching limits or configurational abrupt changes is reduced, improving the safety and continuity of the multi-axis linkage robot's execution process. Furthermore, a comprehensive evaluation function integrating welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost is constructed. Under geometric fidelity constraints, iterative optimization of the key point set is performed, achieving a comprehensive balance between surface fit, motion smoothness, and heat input stability in the generated cladding path, thereby improving the forming quality and processing efficiency of arc cladding operations. Attached Figure Description

[0020] Figure 1 A flowchart of the first embodiment; Figure 2 This is a schematic diagram of the joint angle curve; Figure 3 This is a schematic diagram of the path curvature comparison curves; Figure 4 This is a schematic diagram of bar chart comparison. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the first embodiment, the present invention proposes a path planning method for a multi-axis linkage arc cladding robot, such as... Figure 1 As shown, it includes the following steps: S1. Obtain the point cloud data of the workpiece to be clad and construct an octree structure; calculate the point cloud density and normal vector dispersion within the voxels of the leaf nodes of the octree, and perform initial pruning on voxels whose point cloud density is lower than a first threshold and whose normal vector dispersion is lower than a second threshold as non-critical regions; pass the normal vector dispersion of the child nodes up to the parent node to represent the surface complexity; use the surface complexity of the parent node to adjust the pruning threshold of the child nodes to perform top-down pruning to obtain a simplified point cloud.

[0023] A point cloud dataset of the workpiece to be clad is acquired from a structured light 3D scanner. The bounding boxes of the point cloud are spatially divided according to preset resolution parameters to establish a hierarchical octree structure. At the bottom layer of the octree, i.e., the leaf nodes, the number of point clouds contained in each voxel is counted, and this number is divided by the volume of the voxel to obtain the point cloud density. Principal component analysis is used to calculate the surface normal vector of each point within the voxel. The eigenvalues ​​of the 3D covariance matrix of all normal vectors within the voxel are calculated, and the ratio of the minimum eigenvalue to the sum of all eigenvalues ​​is defined as the normal vector dispersion. The point cloud density and normal vector dispersion of each leaf node are judged. If the point cloud density is lower than a set first threshold and the normal vector dispersion is lower than a set second threshold, the leaf node and its contained point cloud data are treated as non-critical areas and removed in the initial pruning. The first and second thresholds can be preset based on historical experience, the point cloud density distribution of similar workpieces, the statistical results of normal vector dispersion, and the requirements for retaining complex surface features. Traversing the octree hierarchically from bottom to top, the arithmetic mean of the normal vector dispersion of all child nodes belonging to the same parent node is calculated as the surface complexity of that parent node, and this process continues to propagate to higher-level nodes until the root node. Traversing the octree from top to bottom, a linear negative correlation mapping algorithm is used to map the surface complexity of the parent node to the pruning threshold of its child nodes; that is, the higher the surface complexity, the lower the child node density pruning threshold, and vice versa. For example, let the surface complexity of the parent node be... The preset lower limit of surface complexity is The upper limit is The preset lower limit of the child node density pruning threshold is The upper limit is First, calculate the normalized surface complexity. and will Limit the range to 0-1; then calculate the child node density pruning threshold. ,in, This sets the density pruning threshold for child nodes under the current parent node. It makes it easier to preserve point clouds with relatively low density but geometric features in complex curved regions, and to remove redundant low-density point clouds in flat regions.

[0024] Based on the adjusted dynamic point cloud density pruning threshold, the nodes retained after the initial pruning are filtered again from top to bottom. When the node's point cloud density is lower than the corresponding dynamic point cloud density pruning threshold and its normal vector dispersion is lower than the second threshold, the node and its contained point cloud data are deleted. When the node's normal vector dispersion reaches or exceeds the second threshold, the node is retained even if its point cloud density is lower than the corresponding dynamic point cloud density pruning threshold. The point cloud data within all the nodes that are ultimately retained are merged to generate a simplified point cloud.

[0025] In an optional embodiment, acquiring the point cloud data of the workpiece to be clad and constructing an octree structure includes: Scan the surface of the workpiece to be clad to obtain the original three-dimensional point cloud data; Calculate the extreme values ​​of the original 3D point cloud data in the three coordinate axes, generate an initial bounding box that surrounds the original 3D point cloud data, and use the initial bounding box as the root node of the octree; The space corresponding to the root node is divided in half along the midpoint of the three coordinate axes to generate eight subspaces of the same volume as child nodes. The number of point clouds contained in each child node is recursively determined. If the number of point clouds in the child node is greater than a preset threshold and the spatial size of the child node is greater than a preset lower limit, the child node is further divided into equal parts until all leaf nodes satisfy the condition that the number of point clouds is not greater than the threshold or the spatial size reaches the lower limit.

[0026] A non-contact scanning method using a line-structured light 3D laser scanner with an accuracy of 0.01 mm is employed to obtain an original discrete point cloud containing X, Y, and Z 3D coordinates and a set of sampling points for constructing neighborhood relationships. The point cloud data is traversed to extract the extreme values ​​of each axial coordinate, for example, extracting the X-axis extreme value span [0, 500 mm]. Based on this, a 3D cuboid bounding box enclosing the original 3D point cloud data is constructed as the root node with depth Level=0. Using a spatial octree algorithm, the parent node is divided into 8 equally sized child voxel spaces along the geometric center planes of the three orthogonal axes. When recursively performing depth search and subspace segmentation, the preset number threshold is set to 50 to 200, and the lower limit of size is set to 1.0 mm to 5.0 mm. When the total number of point clouds in the current child node exceeds 100 and the shortest side length of the current voxel is greater than 2.0 mm, the Level+1 depth is triggered to continue the equal division until all leaf nodes meet the condition of containing no more than 100 point clouds or the spatial size is reduced to the lower limit of 2.0 mm, thus generating a spatial topology mesh that adapts to the surface curvature details.

[0027] In an optional embodiment, the step of calculating the point cloud density and normal vector dispersion within the voxels at the leaf nodes of the octree includes: Traverse the bottom-level leaf nodes of the octree and count the total number of discrete point clouds contained in each leaf node voxel. The point cloud density of the voxel is obtained by dividing the total number of point clouds within the leaf node voxel by the volume of the voxel. Principal component analysis is used to calculate the local normal vectors of each point cloud within the leaf node voxel. The local normal vectors of all point clouds within the voxel are then uniformly oriented and normalized. The variance of the angle between each local normal vector and the average normal vector within the voxel is calculated, and this variance is used as the normal vector dispersion of the voxel.

[0028] Perform a depth-first traversal of all leaf nodes, count the total number of 3D points N within each child voxel, and divide this total number by the volume of that node, for example, the volume of a cube with a side length of 2.0 mm. , to obtain The voxel point cloud density index is calculated. The K-nearest neighbor algorithm is called, and the search neighborhood K is set to 15 to 30. A covariance matrix is ​​constructed centered on the current voxel point. Principal component analysis is performed to solve for the eigenvector corresponding to the minimum eigenvalue, which is then used as the local initial normal vector for that point. The dot product of vectors is calculated using the origin of the laser scanning viewpoint coordinates. The orientation of all local normal vectors is adjusted to the outside of the workpiece and normalized to a modulus of 1. The arithmetic mean direction vector of the normal vectors of all points within the current voxel is calculated. The inverse cosine function is used to calculate the spatial angle in radians between the local normal vector and the average normal vector of each data point, and the variance of this angle set is calculated. If the variance is close to 0.001, it indicates that the voxel is in a flat region; if the variance is large, it is determined to be a region with large curvature changes. This variance is used as the voxel's normal vector dispersion.

[0029] S2, Based on the simplified point cloud, candidate key points are extracted, the inverse motion solution of the welding gun pose corresponding to each candidate key point is calculated, and a motion cost function inversely proportional to the joint rotation angle distance from the mechanism limit distance is constructed; Under the premise of satisfying the consistency of the inverse solution branches, the solution with the lowest cost function value is selected as the optimal solution, and candidate key points with motion cost or geometric deviation exceeding the threshold are eliminated to obtain a coarse key point set.

[0030] A three-dimensional uniform downsampling operation is performed on the simplified point cloud, and the voxel center point set retained after downsampling is extracted as candidate key points. A standard Denavit-Hartenberg parameter (DH parameter) motion model for the multi-axis linkage arc cladding robot is established, and the inverse motion fast solution algorithm module of the robot operating system's robotic arm navigation architecture is used to solve for multiple possible joint angle sequences for the spatial position and desired normal orientation of each candidate key point. For each inverse motion solution sequence, a penalty function algorithm is used to construct a motion cost function. This cost function calculates the distance between each joint angle and the upper and lower mechanical limit boundaries of that joint, selects the smaller distance as the limit margin of that joint, and then weights and sums the reciprocals of the sum of each joint limit margin and a preset positive bias factor to obtain the motion cost function value, thereby reducing the probability of configurations close to mechanical limits being selected. By comparing multiple inverse solutions for the same candidate keypoint, their pose configuration state word parameters in joint space are extracted. Boolean algebra logic is used to restrict the inverse solutions of adjacent candidate keypoints to have the same pose configuration state word parameters to ensure the consistency of inverse solution branches. If the lowest cost solution at a single point is inconsistent with the configuration state word of an adjacent keypoint, the inverse solution with consistent configuration state word and lower cost function value is selected from the remaining inverse solutions of the same candidate keypoint.

[0031] The optimal inverse solution for a candidate keypoint is selected by using a quicksort algorithm, choosing the joint angle sequence with the lowest motion cost function value among the candidate inverse solutions that satisfy the inverse solution branch consistency constraint. The 3D Euclidean distance between each candidate keypoint and its nearest neighbor in the original point cloud data before simplification is calculated as the geometric deviation. Fixed maximum tolerance thresholds for motion cost and deviation are set. All candidate keypoints are iterated through, and those with motion cost function values ​​greater than the maximum tolerance thresholds or geometric deviations greater than the maximum tolerance thresholds are removed. The remaining candidate keypoints are then allocated to memory space according to their spatial location and topological connectivity, generating a coarsely selected keypoint set.

[0032] In an optional embodiment, the step of solving the inverse motion solution of the welding torch pose corresponding to each candidate key point and constructing a motion cost function inversely proportional to the joint rotation angle distance from the mechanism limit distance includes: Based on the robot's link parameters and motion constraints, multiple sets of selectable joint angle solutions for the target pose corresponding to the current candidate keypoint are obtained by solving the inverse motion matrix. For each joint angle solution, the current calculated angle of each joint of the robot is extracted, the absolute value of the difference between the current calculated angle and the upper and lower limit boundaries of the joint for safe operation is calculated, and the smallest absolute value of the difference is selected as the limit margin of the joint. The weighted sum of the limit margin of each joint and the reciprocal of the preset positive bias factor is used as the motion cost function value of the joint angle solution.

[0033] A multi-axis linkage robot motion model is established based on the DH parameter table, and the Cartesian coordinates (X,Y,Z) and Euler attitude angles of the key points are used. The input is processed by the inverse solver to obtain multiple sets of robot multi-axis joint rotation configuration sequences that are equivalent in the mathematical domain. to For each extracted configuration solution, the difference between the calculated angle of each joint and the corresponding extreme value of the actual safety limit range is calculated. For example, if the limit range of axis 1 of a certain base is [-170°, 170°], and the current calculated solution is 150°, then the absolute difference from the upper boundary is 20°, and the difference from the lower boundary is 320°. The smaller value of the two, 0.349 radians, is selected as the safety limit margin of the current axis. Using the bias factor, such as... The reciprocal of the weighted margin plus the bias factor is summed with preset weights. For example, the weights of axes 1-3 of the upper arm are assigned as 0.25, 0.25, and 0.2, and the weights of axes 4-6 of the wrist are set as 0.1, 0.1, and 0.1. The sum of the reciprocals of the weighted margins of each joint is used as the cost function value. The smaller this value, the further the joint configuration is from the mechanism's limits and the closer it is to the safe operating range, such as... Figure 2 As shown.

[0034] S3. Construct a comprehensive performance evaluation function, which includes the normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost. Using the comprehensive performance evaluation function as the objective, under geometric fidelity constraints, iteratively eliminate and optimize the coarse screening key point set until the comprehensive performance evaluation function value converges to obtain a key point sequence. Perform Cartesian space interpolation on the key point sequence to generate the cladding path.

[0035] The spatial angle between the welding torch posture quaternions corresponding to the optimal inverse solutions of adjacent candidate key points is calculated using quaternion dot multiplication algebraic operations. Combined with the path segment lengths between adjacent key points, the welding torch posture angle change rate sequence is calculated, and the root mean square value or standard deviation of this sequence is taken as the welding torch posture smoothness. The spatial straight-line Euclidean distance between adjacent key points is calculated, and the standard deviation of all node spacing values ​​is taken as the node spacing uniformity. The preset cladding current, cladding voltage, welding torch moving speed, and thermal efficiency coefficients of the control system are extracted. The heat input per unit length corresponding to the path segment of each adjacent key point is calculated, and the dispersion coefficient of all unit length heat input sequence data is taken as the local heat input uniformity. The arithmetic mean of all optimal motion cost function values ​​recorded within the coarse-screened key point set is calculated to obtain the average motion cost. The maximum-minimum normalization algorithm is used to map the evaluation index values ​​of the above four independent dimensions to a continuous numerical range of 0 to 1, and constant weights are assigned to each normalized index, which are then summed to construct a comprehensive performance evaluation function.

[0036] During the global optimization process, a simulated annealing algorithm engine is used for iterative elimination optimization. In each cooling iteration, the algorithm randomly selects and attempts to eliminate a non-first / last key node from the coarsely screened key point set sequence. A non-uniform rational B-spline curve fitting algorithm is used to generate a smooth reconstruction path for the remaining 3D point set after node elimination. KD-tree nearest neighbor search or point-to-curve distance calculation methods are then used to calculate the maximum geometric deviation distance between the reconstructed path and the original simplified point cloud surface. If this maximum geometric deviation distance does not exceed the geometric fidelity upper limit threshold set by the process, the elimination operation is considered to satisfy the spatial constraints, thus triggering a recalculation of the latest comprehensive performance evaluation function value for the remaining point set. If the latest function value is less than the original value before the elimination attempt, or meets the Metropolitan Acceptance Criterion generated by the simulated annealing algorithm, the node is officially confirmed to be eliminated from memory. As the annealing temperature control parameters gradually decrease according to the preset exponential cooling function curve and the above-mentioned random elimination process continues, when the relative change rate of the comprehensive performance evaluation function value in multiple consecutive iterations is less than the preset minimum value of one ten-thousandth, it is determined that the global state has converged, the cooling iteration loop is terminated, and the selected key point sequence is output. Cubic spline interpolation is performed on the Cartesian three-dimensional position coordinates of the key point sequence to generate dense, smooth spatial position interpolation points. Quaternion spherical linear interpolation is performed on the quaternions corresponding to the key points, combined with time parameterization to generate smooth transition attitude interpolation points. The dense spatial position interpolation points and attitude interpolation points are synchronously combined and output according to timestamps to generate a Cartesian spatial continuous cladding trajectory path that can be used to control a multi-axis linkage robot system to perform arc additive manufacturing.

[0037] In an optional embodiment, the construction of the comprehensive performance evaluation function, which includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost, includes: According to the expression Calculate the comprehensive performance evaluation function value; Where E represents the value of the comprehensive performance evaluation function; S, D, H, and K represent the welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost after normalization, respectively. Among them, the smoothness of the welding torch posture is calculated by the standard deviation of the rate of change of the welding torch posture angle between adjacent key points; The uniformity of node spacing is calculated using the standard deviation of the distance between adjacent key points; The uniformity of local heat input is calculated by the dispersion coefficient of the heat input per unit length corresponding to each adjacent key point path segment. The heat input per unit length is determined based on the cladding current, cladding voltage, welding torch moving speed and thermal efficiency coefficient. The average motion cost is calculated by averaging the motion costs of the optimal solutions for all key points in the path sequence; These are the preset weighting coefficients for the four evaluation components.

[0038] The welding torch attitude quaternions corresponding to adjacent key points in the path sequence are read sequentially. The attitude angle between adjacent quaternions is calculated, and the attitude angle is divided by the path segment length between the corresponding adjacent key points to obtain the welding torch attitude angle change rate sequence. Then, the standard deviation of the welding torch attitude angle change rate sequence is calculated as the welding torch attitude smoothness index. The smaller the standard deviation of the welding torch attitude angle change rate sequence, the smaller the fluctuation of the welding torch attitude change rate along the path, and the smoother the welding torch attitude transition; for example, the standard deviation of the attitude angle change rate sequence can be controlled within 0.05 rad / mm as the preferred range.

[0039] Calculate the Euclidean distance between adjacent key points in the path sequence sequentially to obtain the spacing sequence between adjacent key points; then calculate the standard deviation of this spacing sequence as an indicator of node spacing uniformity. For example, if the expected spacing between adjacent points is 3mm, and the standard deviation of the spacing is controlled within 0.1mm, it indicates that the distribution of path nodes is relatively uniform.

[0040] For each path segment formed by adjacent key points, read the corresponding cladding current. cladding voltage welding torch moving speed and thermal efficiency coefficient And calculate according to the formula for heat input per unit length Calculate the heat input per unit length of the i-th path segment. Then, based on the heat input per unit length of all path segments... A heat input sequence is constructed, and the mean and standard deviation of the sequence are calculated. The ratio of the standard deviation to the mean is used as the coefficient of variation of the sequence to obtain the local heat input uniformity index. The smaller the coefficient of variation, the smaller the fluctuation of heat input per unit length between different path segments, and the more stable the heat input distribution during the cladding process.

[0041] Extracting the motion cost of the optimal inverse motion solution corresponding to all key points in the path sequence. and calculate all The arithmetic mean of the two values ​​is used as an indicator of the average cost of movement. The smaller this indicator, the further the robot's overall joint configuration is from the mechanism's limit boundaries, and the better its motion safety.

[0042] After completion and After calculation, the Min-Max normalization method is used to map them to S, D, H, and K respectively, and then the weights are configured as follows: and Calculate the comprehensive evaluation function value E. A lower value indicates better overall performance of the current path in terms of welding torch posture smoothness, node spacing uniformity, local heat input stability, and robot motion safety. For example... Figure 3 As shown, the overall curvature of the path is consistent with the curvature of the original surface, with some fluctuations in some areas, indicating that the planned path has a good geometric fit with the workpiece base surface.

[0043] In an optional embodiment, the step of iteratively eliminating and optimizing the set of coarse screening key points under geometric fidelity constraints, with the comprehensive performance evaluation function as the objective, until the comprehensive performance evaluation function value converges, to obtain the key point sequence, includes: Set the maximum number of iterations; In each iteration, we attempt to temporarily remove a single key point from the current set of coarse screening key points, and recalculate the comprehensive performance evaluation function value after removing the key point. Comparing all temporary removal schemes, under the constraint that the geometric deviation between the path generated after removing the key points and the initial point cloud model is less than the set tolerance, the key points that cause the largest decrease in the comprehensive performance evaluation function value are selected for permanent removal, and the current key point set is updated. Repeat the above steps of temporary removal and permanent elimination until the decrease in the comprehensive performance evaluation function value is less than the preset convergence threshold or the maximum number of iterations is reached, and then take the key point set at this time as the key point sequence.

[0044] Set the maximum number of iterations For example, the iteration can be configured to 500 times or set to 30% of the current keypoint set capacity. In a single iteration, any one keypoint in the current keypoint set is temporarily removed sequentially, and the remaining keypoint set is reconstructed and fitted using a cubic non-uniform rational B-spline curve. Combined with a KD-tree query, the normal distance from the reconstructed trajectory to the original surface is calculated. When the maximum geometric deviation is below the set tolerance, the current removal scheme is considered feasible. Comparing all feasible removal schemes, the recalculated comprehensive performance score E is substituted, and the keypoint that causes the largest decrease in the evaluation function value is extracted and permanently removed, and the current keypoint set is updated. This process is repeated until the absolute difference in the decrease of the E value after M consecutive iterations is less than the preset convergence threshold. or reaching the maximum number of iterations. Output the key point sequence.

[0045] In an optional embodiment, the step of performing Cartesian space interpolation on the keypoint sequence to generate the cladding path includes: Extract the Cartesian space pose of each point in the keypoint sequence; Between adjacent key points, a Cartesian space interpolation algorithm is used to smooth the transition of the Cartesian space pose sequence, generating a Cartesian control trajectory with smooth transition in both position and attitude. Combining the optimal angle sequence of the robot joint space corresponding to each point in the key point sequence, the optimal joint space solution of adjacent key points is used as the initial value or branch constraint for inverse motion solution. The Cartesian control trajectory is mapped to the joint space through the robot inverse motion model. The mapped joint trajectory is then checked for joint limit, inverse solution branch consistency, joint velocity and joint acceleration. A joint control trajectory that meets the preset motion constraints is generated, and a smooth and continuous cladding path is output.

[0046] The three-dimensional Cartesian translation position P, containing X, Y, and Z coordinates, of each point in the keypoint sequence is extracted, along with a quaternion vector Q representing the welding torch posture rotation, where Q includes w, x, y, and z components. For the translational spatial components, a fifth-order polynomial spline equation algorithm is used for interpolation calculation within the control cycle to ensure the continuity of the second-order acceleration of the spatial path. For the welding torch posture rotation components, a spherical linear interpolation module combined with a fifth-order time parameterization function is used for interpolation processing to generate a Cartesian control trajectory with continuous velocity and acceleration. The Cartesian control trajectory is mapped to the robot servo axis system using the selected optimal solution configuration angle sequence as the inverse solution reference starting point and the optimal joint space solutions of adjacent keypoints as branch constraints, through a local Jacobian matrix pseudo-inverse strategy. During the mapping process, joint limits, inverse solution branch consistency, joint velocity, and joint acceleration are verified. When the joint velocity or joint acceleration exceeds a preset allowable threshold, correction is performed using a low-pass filter, time reparameterization, or acceleration limiting smoothing processing to output a smooth and continuous cladding path that satisfies the preset motion constraints.

[0047] The experiment employed a six-axis industrial robot and a 0.01mm precision line structured light 3D laser scanner to scan and verify the path planning of complex curved metal workpieces to be clad. The robot control cycle was set to 8ms, and the initial raw point cloud data contained approximately 800,000 discrete points. The lower limit of the octree segmentation size was set to 2.0mm, and the point cloud quantity threshold was set to 100. Four sets of comparative models were set up: the first set used a basic voxel filtering downsampling combined with a conventional trajectory interpolation planning method; the second set introduced a comprehensive performance evaluation function based on the first set, but did not include motion cost constraints; the third set only used the inverse motion optimal solution selection, without the iterative elimination step of the evaluation function; the fourth set was the complete application of the algorithm of this invention, including octree feature extraction, motion cost constraints, and iterative optimization of the comprehensive evaluation function.

[0048] The results show that the first scheme generates a path with a maximum geometric deviation of 0.85 mm, a minimum margin of 0.04 rad between the joints of the robotic arm and the limit boundary, and a standard deviation of 0.35 mm between adjacent points. The second scheme controls the maximum geometric deviation to around 0.45 mm and improves the standard deviation of the node spacing to 0.12 mm, but the minimum safety limit margin of the joints is still low at only 0.06 rad, and two servo axis overload alarms occurred during the test. The third scheme increases the minimum limit margin of the joints to 0.28 rad, avoiding the limit alarm problem, but the attitude change rate is 0.15 rad / mm, and due to the lack of iterative elimination by the comprehensive evaluation function, the stability of the path node distribution and local thermal input is still insufficient compared to the complete scheme. After 325 iterations of temporary removal and permanent removal, the fourth complete scheme achieved convergence of the comprehensive evaluation function value, stabilized the maximum geometric deviation of the output path within 0.38 mm, reached a minimum joint safety margin of 0.36 rad, optimized the standard deviation of node spacing to 0.08 mm, and reduced the attitude change rate to 0.03 rad / mm. Figure 4 As shown.

[0049] Incorporating the average motion cost into the comprehensive evaluation function, combined with an iterative optimization process, helps the system prioritize joint configurations far from the mechanism's limiting boundaries among multiple inverse motion solutions. This significantly improves the joint safety margin compared to the basic scheme and reduces the risk of robotic arm jitter when approaching limiting or singular regions. The key point iterative elimination mechanism under geometric fidelity constraints reduces redundant points that cause abrupt path changes, ensuring a good curvature fit between the planned path and the original workpiece's base surface, thereby improving the stability of local heat input distribution during the cladding process. Compared to a single optimization strategy, the path after multi-objective weighted fusion exhibits smoother transition characteristics in Cartesian space and achieves a better overall balance between geometric accuracy, motion safety, and path uniformity, improving the forming quality and equipment operational stability of actual cladding operations.

[0050] In a second embodiment, the present invention also provides a path planning system for a multi-axis linkage arc cladding robot, comprising the following modules: The execution module is used to acquire point cloud data of the workpiece to be clad and construct an octree structure; calculate the point cloud density and normal vector dispersion within the voxels in the leaf nodes of the octree, and perform initial pruning on voxels whose point cloud density is lower than a first threshold and whose normal vector dispersion is lower than a second threshold as non-critical regions; pass the normal vector dispersion of the child nodes up to the parent node to represent the surface complexity; and perform top-down pruning by adjusting the pruning threshold of the child nodes using the surface complexity of the parent node to obtain a simplified point cloud. The extraction module is used to extract candidate key points based on the simplified point cloud, calculate the inverse motion solution of the welding torch pose corresponding to each candidate key point, and construct a motion cost function that is inversely proportional to the joint rotation angle distance from the mechanism limit distance. Under the premise of satisfying the consistency of the inverse solution branches, the solution with the lowest cost function value is selected as the optimal solution, and candidate key points with motion cost or geometric deviation exceeding the threshold are eliminated to obtain a coarse key point set. A generation module is used to construct a comprehensive performance evaluation function, which includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost. Using the comprehensive performance evaluation function as the objective, under geometric fidelity constraints, the coarse-screened key point set is iteratively eliminated and optimized until the comprehensive performance evaluation function value converges, resulting in a key point sequence. The key point sequence is then interpolated in Cartesian space to generate the cladding path.

[0051] In this specification, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0052] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0053] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A path planning method for a multi-axis linkage arc cladding robot, characterized in that, Includes the following steps: Obtain point cloud data of the workpiece to be clad and construct an octree structure; calculate the point cloud density and normal vector dispersion within the voxels of the leaf nodes of the octree, and perform initial pruning on voxels whose point cloud density is lower than a first threshold and whose normal vector dispersion is lower than a second threshold as non-critical regions; pass the normal vector dispersion of the child nodes up to the parent node to represent the surface complexity; use the surface complexity of the parent node to adjust the pruning threshold of the child nodes to perform top-down pruning to obtain a simplified point cloud; Based on the simplified point cloud, candidate key points are extracted, the inverse motion solution of the welding gun pose corresponding to each candidate key point is calculated, and a motion cost function inversely proportional to the joint rotation angle distance from the mechanism limit distance is constructed. Under the premise of satisfying the consistency of the inverse solution branches, the solution with the lowest cost function value is selected as the optimal solution, and candidate key points with motion cost or geometric deviation exceeding the threshold are eliminated to obtain a coarse key point set. A comprehensive performance evaluation function is constructed, which includes the normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost. With the comprehensive performance evaluation function as the objective, under the geometric fidelity constraint, the set of key points for coarse screening is iteratively eliminated and optimized until the comprehensive performance evaluation function value converges, thus obtaining the key point sequence. Cartesian space interpolation is performed on the key point sequence to generate the cladding path.

2. The method according to claim 1, characterized in that, The step of acquiring point cloud data of the workpiece to be clad and constructing an octree structure includes: Scan the surface of the workpiece to be clad to obtain the original three-dimensional point cloud data; Calculate the extreme values ​​of the original 3D point cloud data in the three coordinate axes, generate an initial bounding box that surrounds the original 3D point cloud data, and use the initial bounding box as the root node of the octree; The space corresponding to the root node is divided in half along the midpoint of the three coordinate axes to generate eight subspaces of the same volume as child nodes. The number of point clouds contained in each child node is recursively determined. If the number of point clouds in the child node is greater than a preset threshold and the spatial size of the child node is greater than a preset lower limit, the child node is further divided into equal parts until all leaf nodes satisfy the condition that the number of point clouds is not greater than the threshold or the spatial size reaches the lower limit.

3. The method according to claim 1, characterized in that, The calculation of point cloud density and normal vector dispersion within the leaf nodes of the octree includes: Traverse the bottom-level leaf nodes of the octree and count the total number of discrete point clouds contained in each leaf node voxel. The point cloud density of the voxel is obtained by dividing the total number of point clouds within the leaf node voxel by the volume of the voxel. Principal component analysis is used to calculate the local normal vectors of each point cloud within the leaf node voxel. The local normal vectors of all point clouds within the voxel are then uniformly oriented and normalized. The variance of the angle between each local normal vector and the average normal vector within the voxel is calculated, and this variance is used as the normal vector dispersion of the voxel.

4. The method according to claim 1, characterized in that, The process of solving the inverse motion solution of the welding torch pose corresponding to each candidate key point and constructing a motion cost function inversely proportional to the joint rotation angle distance from the mechanism limit distance includes: Based on the robot's link parameters and motion constraints, multiple sets of selectable joint angle solutions for the target pose corresponding to the current candidate keypoint are obtained by solving the inverse motion matrix. For each joint angle solution, the current calculated angle of each joint of the robot is extracted, the absolute value of the difference between the current calculated angle and the upper and lower limit boundaries of the joint for safe operation is calculated, and the smallest absolute value of the difference is selected as the limit margin of the joint. The weighted sum of the limit margin of each joint and the reciprocal of the preset positive bias factor is used as the motion cost function value of the joint angle solution.

5. The method according to claim 2 or 3, characterized in that, The constructed comprehensive performance evaluation function includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost, including: According to the expression Calculate the comprehensive performance evaluation function value; Where E represents the value of the comprehensive performance evaluation function; S, D, H, and K represent the welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost after normalization, respectively. Among them, the smoothness of the welding torch posture is calculated by the standard deviation of the rate of change of the welding torch posture angle between adjacent key points; The uniformity of node spacing is calculated using the standard deviation of the distance between adjacent key points; The uniformity of local heat input is calculated by the dispersion coefficient of the heat input per unit length corresponding to each adjacent key point path segment. The heat input per unit length is determined based on the cladding current, cladding voltage, welding torch moving speed and thermal efficiency coefficient. The average motion cost is calculated by averaging the motion costs of the optimal solutions for all key points in the path sequence; These are the preset weighting coefficients for the four evaluation components.

6. The method according to claim 1, characterized in that, The process involves iteratively eliminating and optimizing the set of key points in the coarse screening, with the comprehensive performance evaluation function as the objective and under geometric fidelity constraints, until the comprehensive performance evaluation function converges, resulting in a key point sequence, including: Set the maximum number of iterations; In each iteration, we attempt to temporarily remove a single key point from the current set of coarse screening key points, and recalculate the comprehensive performance evaluation function value after removing the key point. Comparing all temporary removal schemes, under the constraint that the geometric deviation between the path generated after removing the key points and the initial point cloud model is less than the set tolerance, the key points that cause the largest decrease in the comprehensive performance evaluation function value are selected for permanent removal, and the current key point set is updated. Repeat the above steps of temporary removal and permanent elimination until the decrease in the comprehensive performance evaluation function value is less than the preset convergence threshold or the maximum number of iterations is reached, and then take the key point set at this time as the key point sequence.

7. The method according to claim 1, characterized in that, The step of performing Cartesian space interpolation on the key point sequence to generate the cladding path includes: Extract the Cartesian space pose of each point in the keypoint sequence; Between adjacent key points, a Cartesian space interpolation algorithm is used to smooth the transition of the Cartesian space pose sequence, generating a Cartesian control trajectory with smooth transition in both position and attitude. Combining the optimal angle sequence of the robot joint space corresponding to each point in the key point sequence, the optimal joint space solution of adjacent key points is used as the initial value or branch constraint for inverse motion solution. The Cartesian control trajectory is mapped to the joint space through the robot inverse motion model. The mapped joint trajectory is then checked for joint limit, inverse solution branch consistency, joint velocity and joint acceleration. A joint control trajectory that meets the preset motion constraints is generated, and a smooth and continuous cladding path is output.

8. A path planning system for a multi-axis linkage arc cladding robot, characterized in that, Includes the following modules: The execution module is used to acquire point cloud data of the workpiece to be clad and construct an octree structure; calculate the point cloud density and normal vector dispersion within the voxels in the leaf nodes of the octree, and perform initial pruning on voxels whose point cloud density is lower than a first threshold and whose normal vector dispersion is lower than a second threshold as non-critical regions; pass the normal vector dispersion of the child nodes up to the parent node to represent the surface complexity; and perform top-down pruning by adjusting the pruning threshold of the child nodes using the surface complexity of the parent node to obtain a simplified point cloud. The extraction module is used to extract candidate key points based on the simplified point cloud, calculate the inverse motion solution of the welding torch pose corresponding to each candidate key point, and construct a motion cost function that is inversely proportional to the joint rotation angle distance from the mechanism limit distance. Under the premise of satisfying the consistency of the inverse solution branches, the solution with the lowest cost function value is selected as the optimal solution, and candidate key points with motion cost or geometric deviation exceeding the threshold are eliminated to obtain a coarse key point set. A generation module is used to construct a comprehensive performance evaluation function, which includes normalized welding torch posture smoothness, node spacing uniformity, local heat input uniformity, and average motion cost. With the comprehensive performance evaluation function as the objective, under the geometric fidelity constraint, the set of key points for coarse screening is iteratively eliminated and optimized until the comprehensive performance evaluation function value converges, thus obtaining the key point sequence. Cartesian space interpolation is performed on the key point sequence to generate the cladding path.

9. The system according to claim 8, characterized in that, The step of acquiring point cloud data of the workpiece to be clad and constructing an octree structure includes: Scan the surface of the workpiece to be clad to obtain the original three-dimensional point cloud data; Calculate the extreme values ​​of the original 3D point cloud data in the three coordinate axes, generate an initial bounding box that surrounds the original 3D point cloud data, and use the initial bounding box as the root node of the octree; The space corresponding to the root node is divided in half along the midpoint of the three coordinate axes to generate eight subspaces of the same volume as child nodes. The number of point clouds contained in each child node is recursively determined. If the number of point clouds in the child node is greater than a preset threshold and the spatial size of the child node is greater than a preset lower limit, the child node is further divided into equal parts until all leaf nodes satisfy the condition that the number of point clouds is not greater than the threshold or the spatial size reaches the lower limit.

10. The system according to claim 8, characterized in that, The calculation of point cloud density and normal vector dispersion within the leaf nodes of the octree includes: Traverse the bottom-level leaf nodes of the octree and count the total number of discrete point clouds contained in each leaf node voxel. The point cloud density of the voxel is obtained by dividing the total number of point clouds within the leaf node voxel by the volume of the voxel. Principal component analysis is used to calculate the local normal vectors of each point cloud within the leaf node voxel. The local normal vectors of all point clouds within the voxel are then uniformly oriented and normalized. The variance of the angle between each local normal vector and the average normal vector within the voxel is calculated, and this variance is used as the normal vector dispersion of the voxel.

Images