Robot flexible welding path planning method based on dynamic batch sampling optimization

Abstract

The present invention discloses a method for robot flexible welding path planning based on dynamic batch sampling optimization. The method comprises the following steps: first, acquiring and processing point cloud information of the welding workpiece surface to obtain a three-dimensional model of the welding workpiece; then, applying a feature recognition algorithm to complete weld extraction; then, determining the weld seam posture and welding gun posture; applying a meta-heuristic algorithm to determine the welding sequence of the welds; finally, applying a dynamic path generation algorithm to quickly find a collision-free path to complete welding path planning; and utilizing the planned welding path parameters to guide the welding robot to perform welding. The present invention applies a dynamic path generation algorithm to dynamically batch sample nodes in the welding robot's joint space to find a collision-free path for the welding robot, which can improve planning efficiency and path quality in solving complex constraint path planning problems.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot flexible welding and relates to a robot flexible welding path planning method, in particular to a robot flexible welding path planning method based on dynamic batch sampling optimization. Background Art

[0002] With the in-depth development of industrial automation, industrial robots have been widely used in standardized and streamlined operations in industries such as the automotive industry, electronic assembly, metallurgy and chemical industry, and food processing. They are mainly used for operations such as grinding, painting, loading and unloading, and welding. At present, the welding tasks of large-scale standardized products have been automated through manual teaching programming. However, in fields such as shipbuilding, offshore engineering, and steel structures, the diversity and degree of non-standardization of production make it difficult to compile unified welding programs to meet the flexible welding needs of small-batch workpieces with multiple specifications. Traditional manual teaching methods have many drawbacks, such as long design cycles, low production efficiency, and high labor costs. Therefore, for such personalized, multi-specification, small-batch workpieces, it is necessary to introduce robotic flexible welding technology and develop a more automated and intelligent robotic welding method.

[0003] Robotic flexible welding is an advanced welding technology that uses multi-axis industrial robot operation. It can flexibly configure corresponding data processing algorithms according to different welding processes and requirements, extract processing information and control the robot to complete a series of welding tasks. This technology can significantly improve the automation level and efficiency of welding operations, and achieve efficient, high-quality, high-precision and flexible automatic welding of small batches of workpieces of multiple specifications. Related research on robotic flexible welding technology mainly focuses on weld starting position guidance, weld extraction, weld tracking, welding pool monitoring and welding path planning. Among them, welding path planning technology is particularly critical, requiring the robot to plan a collision-free and optimized welding gun movement path based on the identified workpiece information and weld information. Existing flexible welding path planning methods mainly include:

[0004] 1) Parameter-based path planning. This method uses a predefined set of parameters, such as gun speed, gun angle, and gun-to-workpiece distance, and adjusts these parameters to generate a path tailored to the specific welding task. This method lacks flexibility, especially when processing complex or non-standard workpieces. Parameter adjustment is difficult, and achieving the desired welding result is difficult.

[0005] 2) Path planning based on a gridded map. This method first divides the weld area into a grid to create a gridded map. Graph search algorithms such as D*Lit, Theta*, Dijkstra, and A* are then used to find a solution path on this map. The accuracy of this path planning is limited by the resolution of the grid, so creating the gridded map requires significant time and computational effort, making it unsuitable for production of small batches of workpieces with multiple specifications and rapidly changing requirements.

[0006] 3) Path planning based on sensor feedback. This method relies on real-time sensor feedback to dynamically optimize and adjust the welding path. This method can significantly improve welding process adaptability and accuracy when processing complex or non-standard workpieces. However, this method is highly dependent on sensor accuracy and response speed, has high requirements for the welding environment, and faces frequent changes in workpiece types and welding requirements, requiring constant adjustment and calibration of the sensor system, which is costly. Summary of the Invention

[0007] In the process of completing welding automation for the production of small batches of workpieces with multiple specifications, in order to solve the defects existing in the background technology, the present invention provides a robot flexible welding path planning method based on dynamic batch sampling optimization. The present invention applies a dynamic path generation algorithm to perform dynamic batch sampling nodes in the joint space of the welding robot to find a collision-free path for the welding robot, which can improve planning efficiency and path quality in solving path planning problems with complex constraints.

[0008] The technical solutions of the present invention are as follows:

[0009] 1. A robot flexible welding path planning method based on dynamic batch sampling optimization

[0010] Step 1: After acquiring and processing the point cloud information of the welding workpiece surface, a three-dimensional model of the welding workpiece is obtained;

[0011] Step 2: Extract the weld seam from the three-dimensional model of the workpiece according to the size information of the welded workpiece, and then determine the welding sequence of the weld seam;

[0012] Step 3: Determine the weld posture and welding gun posture based on the 3D model of the welded workpiece and the weld;

[0013] Step 4: Use the directed bounding box method to determine the obstacle objects and then establish the welding environment model;

[0014] Step 5: Based on the weld posture, welding gun posture and welding environment model, combined with the welding sequence of the weld, a dynamic path generation algorithm is applied to quickly find a collision-free path and obtain the welding path;

[0015] Step 6: Send the welding path parameters to the robot control cabinet, and then control the welding robot to weld the workpiece.

[0016] In the second step, a meta-heuristic algorithm is used to determine the welding sequence of the welds according to the welds of the welded workpiece.

[0017] In step 5, the welding path includes the planned path between the start and end points of the weld and the start or stop motion state of the robot and the planned path between two separated welds.

[0018] In step 5, the specific process of using the dynamic path generation algorithm to quickly find a collision-free path based on the weld posture, welding gun posture and welding environment model is as follows:

[0019] S1: S1: Establish the robot path node set N and the robot section set W, and set the robot arm starting state node x start Put it into the robot path node set N;

[0020] S2: Based on the weld posture, welding gun posture and welding environment model, combined with the current robot path node set N, the robot joint space is discretized into a robot sampling node set, which contains the robot arm target state node x end and the robot's initial state node x start , continuously obtain the robot path nodes and the corresponding robot road segments in the robot sampling node set, and put the obtained robot path nodes into the robot path node set N, and the corresponding robot road segments into the robot road segment set W, until the robot arm target state node x end Put it into the robot path node set N to obtain the initial planning path;

[0021] S3: Dynamically optimize the initial planned path to obtain the optimal planned path.

[0022] The S2 is specifically:

[0023] S2.1: Randomly sample a set of nodes from the robot sampling node set as potential path nodes and add them to the path node queue Q N In this group of nodes, the robot target state node x is included. end and the robot's initial state node x start ;

[0024] S2.2: For the path node queue Q N Each potential path node x in extend , establish the current potential path node x extend and each neighbor node x in its neighborhood D neighbor The potential connection between them, traversing the path node queue QN Each potential path node x extend , thus obtaining the path node queue Q N All potential connections in

[0025] S2.3: Use the three-dimensional model of the welding workpiece to calculate the path node queue Q N All potential connections in the collision detection are performed separately, and the potential connections that pass the collision detection are regarded as candidate links and added to the link queue Q w middle;

[0026] S2.4: Calculate the road segment queue Q based on the cost evaluation function C w The cost of each candidate segment in the robot segment set W is added to the candidate segment with the lowest cost, and the neighbor node x corresponding to the candidate segment is added to the robot segment set W. neighbor Add to the robot path node set N; and clear the path node queue Q N and segment queue Q w ;

[0027] S2.5: Repeat S2.1 to S2.4 until the robot reaches the target state node x end As a neighbor node x neighbor The robot is added to the path node set N, and the initial planned path is obtained by combining the robot road segment set W and the robot path node set N.

[0028] The S3 is specifically:

[0029] S3.1: Construct a measurement matrix M based on the initial planned path. After performing singular value decomposition and calculation on the measurement matrix M, the super ellipsoid rotation matrix C is obtained. The formula is as follows:

[0030]

[0031] U∑V T ≡M

[0032] C=Udiag{1,...,det(U)det(V)}V T

[0033] Among them, a1 represents the starting point x of the current planned path start To the end point x of the initial planned path end direction, U,∑,V T are the left singular vector matrix, diagonal matrix and right singular vector transposed matrix respectively, || ||2 represents the Euclidean norm; I1 is the 1*n dimensional identity matrix, T represents the transpose, det() represents the determinant of the matrix; diag{} represents the diagonal matrix;

[0034] S3.2: Based on the length of the current planned path and the starting point x start and the end point x end To establish a super ellipsoid, the formulas for related geometric parameters are as follows:

[0035]

[0036] Among them, a, b, and c represent the major axis, minor axis, and focal length of the super ellipsoid, respectively, and c best Represents the current planned path length, c min Represents the starting point x of the current planned path start and the end point x end the distance between them;

[0037] S3.3: Construct the shape matrix L of the superellipsoid, as follows:

[0038]

[0039] S3.4: Randomly sample a set of nodes x within the unit hypersphere ball , the node x is rotated by the shape matrix L and the super ellipsoid rotation matrix C ball After scaling, rotating and translating, a sampling node group A within the super ellipsoid is obtained;

[0040] S3.5: At each current path point x way Generate a corresponding set of random sampling nodes B in the neighborhood K of , and then connect the sampling node group A with each point x way After the sampling node group B is fused, the optimized sampling node set S is obtained;

[0041] S3.6: Based on the robot arm target state node x end and the robot's initial state node x start , combined with the current sampling node optimization set S, generate the optimized planning path;

[0042] S3.7: Repeat S3.1 to S3.6 until the optimal planned path is obtained.

[0043] The S3.6 is specifically:

[0044] S3.6.1: Initialize the robot path node set N′ and the robot segment set W′, and set the robot arm starting state node x start Put it into the robot path node set N′;

[0045] S3.6.2: Randomly select a set of nodes from the sampling node optimization set S as potential path nodes and add them to the path node queue Q′ N In this group of nodes, the target state node x of the robot arm must be included. end and the robot's initial state node xstart ;

[0046] S3.6.3: For the path node queue Q′ N Each potential path node x′ in extend , establish the current potential path node x′ extend and each neighbor node x′ in its neighborhood D neighbor The potential connection between them, traversing the path node queue Q′ N Each potential path node x′ extend , thus obtaining the path node queue Q′ N All potential connections in

[0047] S3.6.4: Use the three-dimensional model of the welding workpiece to calculate the path node queue Q′ N All potential connections in the collision detection are performed separately, and the potential connections that pass the collision detection are regarded as candidate links and added to the link queue Q′ w middle;

[0048] S3.6.5: Calculate the road segment queue Q′ based on the cost evaluation function C w The cost of each candidate segment in the robot is added to the robot segment set W′ and the neighbor node x′ corresponding to the candidate segment is calculated. neighbor Add to the robot path node set N' and clear the path node queue Q' N and the road segment queue Q′ w ;

[0049] S3.6.6: Repeat S3.6.1 to S3.6.5 until the robot reaches the target state node x end As a neighbor node x′ neighbor The robot path node set N′ is added, and the optimized planning path is obtained by combining the robot road segment set W′ and the robot path node set N′.

[0050] 2. A computer device

[0051] The device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method when executing the computer program.

[0052] 3. A computer-readable storage medium

[0053] The medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0054] 4. A computer program product

[0055] The product comprises a computer program / instructions which, when executed by a processor, implement the steps of the method.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. The present invention can flexibly configure corresponding data processing algorithms according to different welding processes and requirements, extract processing information and control the robot to complete a series of welding tasks. It is universal for workpieces of multiple specifications and small batches, solving the problems of insufficient flexibility and low production efficiency of existing technologies when dealing with diversified and non-standard products.

[0058] 2. This paper proposes a dynamic path generation algorithm that uses batch point sampling in the welding robot's joint space and dynamically optimizes the initial path after generation. This algorithm effectively improves path quality and reduces planning time, while ensuring continuity of the welding operation and consistent weld quality.

[0059] 3. The dynamic path generation algorithm proposed in this paper optimizes the initial solution path by using a constrained sampling space and increasing sampling around the initial robot path nodes. Constraining the sampling space avoids a large number of ineffective detections, saving time and computational costs. Increasing sampling around the initial robot path nodes allows the robot to move as close to the workpiece edge as possible. By combining these two sampling options, path quality is optimized.

[0060] 4. This invention proposes a flexible robotic welding method based on dynamic batch sampling optimization. This method effectively incorporates advanced automation technologies and data processing methods, such as point cloud 3D reconstruction and weld feature extraction, before applying the proposed efficient path planning method. This significantly improves the automation and integration of the welding process, promoting the development of intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a general flow chart of the robot flexible welding path planning method provided by an embodiment of the present invention.

[0062] Figure 2 It is a flexible automatic welding system.

[0063] Figure 3 This is a flowchart of the initial solution path obtained by applying the dynamic path generation algorithm in the example of the present invention.

[0064] Figure 4 This is a flowchart of the optimal path obtained by dynamic optimization based on the initial solution path in the example of the present invention.

[0065] In the figure: 1. Welding robot; 2. Turntable; 3. Welding workpiece; 4. 3D structured light camera; 5. Scanning frame; 6. Host computer; 7. Welding power supply; 8. Robot control cabinet. DETAILED DESCRIPTION

[0066] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be noted that the embodiments described below are intended to facilitate understanding of the present invention and do not have any limiting effect on the present invention.

[0067] As attached Figure 2 As shown, a flexible automatic welding system includes: a welding robot 1, a turntable 2, a welding workpiece 3, a three-dimensional structured light camera 4, a scanning frame 5, a host computer 6, a welding power supply 7 and a robot control cabinet 8.

[0068] The host computer 6 receives the workpiece surface point cloud information captured by the 3D structured light camera 4, performs point cloud stitching and workpiece surface reconstruction, completes weld seam extraction, and plans the welding path for the welding robot 3. The welding power supply 7 provides power to the robot control cabinet 8, the welding robot 1, the turntable 2, and the 3D structured light camera 4. The robot control cabinet 8 receives the welding path, trajectory parameters, and welding process parameters from the host computer 6, and controls the welding process of the welding robot 1 and the mechanical movement of the turntable 2. The workpiece 3 is placed on the turntable 2, and the camera's height and angle can be adjusted by adjusting the position of the hinge.

[0069] This paper proposes a dynamic path generation algorithm that enables the robot to quickly find the optimal collision-free path;

[0070] like Figure 1 As shown, a robot flexible welding path planning method based on dynamic batch sampling optimization includes the following steps:

[0071] Step 1: After the flexible automatic welding system obtains and processes the surface point cloud information of the welding workpiece, the point cloud processing specifically performs point cloud stitching and three-dimensional reconstruction of the workpiece surface to obtain a three-dimensional model of the welding workpiece; in this embodiment, the turntable 2 cooperates with the three-dimensional structured light camera 4 to obtain the surface point cloud information of the welding workpiece 3, and transmits the information to the host computer 1; the host computer 6 receives the surface point cloud information of the workpiece taken by the three-dimensional structured light camera 4, and performs point cloud stitching and three-dimensional reconstruction of the workpiece surface.

[0072] Step 2: Based on the size information of the welded workpiece, the feature recognition algorithm is used to extract the weld seam from the three-dimensional model of the workpiece to obtain the weld seam of the welded workpiece; then the welding sequence of the weld seam is determined;

[0073] In step 2, the welding sequence of the welds is determined using a meta-heuristic algorithm according to the welds of the welded workpiece.

[0074] Step 3: Determine the weld posture and welding gun posture according to the three-dimensional model of the welding workpiece and the weld. The weld posture is the weld posture spatial orientation, and the welding gun posture is the welding gun posture spatial orientation.

[0075] Step 4: Use the directed bounding box method to determine the obstacle objects and then establish the welding environment model;

[0076] Step 5: Based on the weld seam posture, welding gun posture, and welding environment model (such as 3D point cloud coordinate information of obstacles), combined with the welding sequence, a dynamic path generation algorithm is applied to quickly find a collision-free path to obtain the welding path. The welding path includes the planned path between the weld start and end points and the robot's start or stop state, as well as the planned path between two separate weld segments. The weld start and end points are specifically the starting point of the first weld and the end point of the last weld.

[0077] Specifically:

[0078] The specific process of using the dynamic path generation algorithm to quickly find a collision-free path based on the weld posture, welding gun posture, and welding environment model is as follows:

[0079] S1: Establish the robot path node set N and the robot section set W, and set the robot arm starting state node x start Put it into the robot path node set N, and the robot segment set W is empty. Create segment queue Q w and path node queue Q N , set the limit step size S limit 、W limit And the number of iterations. The robot arm starts at the state node x start Contains the robot's starting state point, the starting point of two adjacent separation welds or the end point of the weld, and the robot's target state node x end It includes the starting and ending points of the weld, the end points of two adjacent separated welds, or the robot stop state point.

[0080] S2: Based on the weld posture, welding gun posture and welding environment model, combined with the current robot path node set N, the robot joint space is discretized into a robot sampling node set, which contains the robot arm target state node x end and the robot's initial state node x start , continuously obtain the robot path nodes and the corresponding robot road segments in the robot sampling node set, and put the obtained robot path nodes into the robot path node set N, and the corresponding robot road segments into the robot road segment set W, until the robot arm target state node x end Put it into the robot path node set N to obtain the initial planning path;

[0081] S2 is specifically:

[0082] S2.1: Randomly sample a set of nodes from the robot sampling node set as potential path nodes and add them to the path node queue Q N In this group of nodes, the robot target state node x is included. end and the robot's initial state node x start ;

[0083] S2.2: For the path node queue Q N Each potential path node x in extend , establish the current potential path node x extend and each neighbor node x in its neighborhood D neighbor The potential connection between them, traversing the path node queue Q N Each potential path node x extend , thus obtaining the path node queue Q N All potential connections in

[0084] Neighborhood D is a hypersphere neighborhood that includes the distance x in the path graph. extend No more than S limit For all nodes, the formula is as follows:

[0085] D(x extend )={x in joint space: dis(x,x extend )≤S limit}

[0086] Among them, D(x extend ) is x extend The hypersphere neighborhood of dis(x,x extend ) is x extend The distance between S and x; limit To set the neighborhood radius;

[0087] S2.3: Using the 3D model of the welding workpiece, that is, the 3D reconstructed workpiece point cloud data, the path node queue Q is detected by the PCL point cloud collision detection library. N All potential connections in the collision detection are performed separately, and the potential connections that pass the collision detection are regarded as candidate links and added to the link queue Q w In the middle; the potential connection that does not pass the collision detection is moved to the connection blacklist, and it no longer exists as a potential connection;

[0088] S2.4: Calculate the road segment queue Q based on the cost evaluation function C w The cost of each candidate segment in the robot segment set W is added to the candidate segment with the lowest cost, and the neighbor node x corresponding to the candidate segment is added to the robot segment set W. neighbor Add to the robot path node set N; and clear the path node queue QN and segment queue Q w ,

[0089] The calculation formula of the cost evaluation function C is as follows:

[0090] C=C1+C2

[0091]

[0092] Among them, C2 represents the distance from the current evaluation node to the target node x end The cost, θ n,i Represents the angle of the i-th joint in the current state, θ end,i Represents the angle of the i-th joint in the target state.

[0093] S2.5: Repeat S2.1 to S2.4 until the robot reaches the target state node x end As a neighbor node x neigh The robot is added to the path node set N, and the initial planned path is obtained by combining the robot road segment set W and the robot path node set N.

[0094] S3: Dynamically optimize the initial planned path to obtain the optimal planned path, thereby obtaining the welding path.

[0095] S3 specifically:

[0096] S3.1: Construct a measurement matrix M based on the current planned path. After performing singular value decomposition and calculation on the measurement matrix M, the superellipsoid rotation matrix C is obtained. The formula is as follows:

[0097]

[0098] U∑V T ≡M

[0099] C=Udiag{1,...,det(U)det(V)}V T

[0100] Among them, a1 is a unit vector, representing the starting point x of the current planned path start To the end point x of the initial planned path end direction, U,∑,V T Represents the superellipsoid transformation matrix, which are the left singular vector matrix, diagonal matrix and right singular vector transposed matrix respectively, || ||2 represents the Euclidean norm; I1 is the 1*n dimensional identity matrix, T represents the transpose, det() represents the determinant of the matrix; diag{} represents the diagonal matrix;

[0101] S3.2: Based on the length of the current planned path and the starting point x start and the end point xend To establish a super ellipsoid, the formulas for related geometric parameters are as follows:

[0102]

[0103] Among them, a, b, and c represent the major axis, minor axis, and focal length of the super ellipsoid, respectively, and c best Represents the current planned path length, c min Represents the starting point x of the current planned path start and the end point x end the distance between them;

[0104] S3.3: Construct the shape matrix L of the super ellipsoid. The shape matrix L is a diagonal matrix based on c best and c min The elements on the diagonal are defined to represent the axis lengths of the superellipsoid in each dimension, which determines the shape of the superellipsoid in high-dimensional space; the formula is as follows:

[0105]

[0106] S3.4: Randomly sample a set of nodes x within the unit hypersphere ball , the node x is rotated by the shape matrix L and the super ellipsoid rotation matrix C ball After scaling, rotating and translating, a sampling node group A within the super ellipsoid is obtained;

[0107] Each sampling node in sampling node group A The formula for obtaining is:

[0108] x ball ~u(x ball )

[0109]

[0110] Where u( ) represents a function randomly sampled within a unit hypersphere, x centre Represents the center point of the superellipse, which is the midpoint between the two foci.

[0111] S3.5: At each current path point x way Generate a corresponding set of random sampling nodes B in the neighborhood K of each path point of the current planned path, and then compare the sampling node group A with each point x way After the sampling node group B is fused, the optimized sampling node set S is obtained;

[0112] S3.6: Based on the robot arm target state node x end and the robot's initial state node x start , combined with the current sampling node optimization set S, generate the optimized planning path;

[0113] S3.6 specifically reads:

[0114] S3.6.1: Initialize the robot path node set N′ and the robot segment set W′, and set the robot arm starting state node x start Put it into the robot path node set N', the robot segment set W' is empty. Create segment queue Q' w and path node queue Q′ N , set the limit step size S limit 、W limit And the number of iterations. The robot arm starts at the state node x start Contains the robot's starting state point, the starting point of two adjacent separation welds or the end point of the weld, and the robot's target state node x end It includes the starting and ending points of the weld, the end points of two adjacent separated welds, or the robot stop state point.

[0115] S3.6.2: Randomly select a set of nodes from the sampling node optimization set S as potential path nodes and add them to the path node queue Q′ N In this group of nodes, the target state node x of the robot arm must be included. end and the robot's initial state node x start ;

[0116] S3.6.3: For the path node queue Q′ N Each potential path node x′ in extend , establish the current potential path node x′ extend and each neighbor node x′ in its neighborhood D nexighbor The potential connection between them, traversing the path node queue Q′ N Each potential path node x′ extend , thus obtaining the path node queue Q′ N All potential connections in

[0117] S3.6.4: Use the three-dimensional model of the welding workpiece to calculate the path node queue Q′ N All potential connections in the collision detection are performed separately, and the potential connections that pass the collision detection are regarded as candidate links and added to the link queue Q′ w middle;

[0118] S3.6.5: Calculate the road segment queue Q′ based on the cost evaluation function C w The cost of each candidate segment in the robot is added to the robot segment set W′ and the neighbor node x′ corresponding to the candidate segment is calculated. neighbor Add to the robot path node set N' and clear the path node queue Q' N and the road segment queue Q′ w ;

[0119] S3.6.6: Repeat S3.6.1 to S3.6.5 until the robot reaches the target state node x end As a neighbor node x′ neighbo The robot path node set N′ is added, and the optimized planning path is obtained by combining the robot road segment set W′ and the robot path node set N′.

[0120] S3.7: Repeat S3.1 to S3.6 until the set number of iterations is reached (one execution is equivalent to one optimization) to obtain the optimal planning path.

[0121] Step 6: Send the welding path parameters to the robot control cabinet 8, and then control the welding robot 1 to weld the welding workpiece.

[0122] Finally, it should be noted that the above embodiments and explanations are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. It should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention may be made without departing from the spirit and scope of the technical solutions disclosed herein, and all such modifications or equivalent substitutions shall be encompassed within the scope of protection of the claims of the present invention.

Claims

1. A robot flexible welding path planning method based on dynamic batch sampling optimization, characterized in that: The following steps are involved: Step 1: After acquiring and processing the point cloud information of the welding workpiece surface, a three-dimensional model of the welding workpiece is obtained; Step 2: Extract the weld seam from the three-dimensional model of the workpiece based on the size information of the welded workpiece to obtain the weld seam of the welded workpiece; and then determine the welding sequence of the weld seam; Step 3: Determine the weld posture and welding gun posture based on the 3D model of the welded workpiece and the weld; Step 4: Use the directed bounding box method to determine the obstacle objects and then establish the welding environment model; Step 5: Based on the weld posture, welding gun posture and welding environment model, combined with the welding sequence of the weld, a dynamic path generation algorithm is applied to quickly find a collision-free path and obtain the welding path; In step 5, the specific process of using the dynamic path generation algorithm to quickly find a collision-free path based on the weld posture, welding gun posture and welding environment model is as follows: S1: Establish the robot path node set N and the robot section set W, and set the robot arm starting state node x start Put it into the robot path node set N; S2: Based on the weld posture, welding gun posture and welding environment model, combined with the current robot path node set N, the robot joint space is discretized into a robot sampling node set, which contains the robot arm target state node x end and the robot's initial state node x start , continuously obtain the robot path nodes and the corresponding robot road segments in the robot sampling node set, and put the obtained robot path nodes into the robot path node set N, and the corresponding robot road segments into the robot road segment set W, until the robot arm target state node x end Put it into the robot path node set N to obtain the initial planning path; S3: Dynamically optimize the initial planned path to obtain the optimal planned path; The S2 is specifically: S2.1: Randomly sample a set of nodes from the robot sampling node set as potential path nodes and add them to the path node queue Q N In this group of nodes, the robot target state node x is included. end and the robot's initial state node x start ; S2.2: For the path node queue Q N Each potential path node x in extend , establish the current potential path node x extend and each neighbor node x in its neighborhood D neighbor The potential connection between them, traversing the path node queue Q N Each potential path node x extend , thus obtaining the path node queue Q N All potential connections in S2.3: Use the three-dimensional model of the welding workpiece to calculate the path node queue Q N All potential connections in the collision detection are performed separately, and the potential connections that pass the collision detection are regarded as candidate links and added to the link queue Q w middle; S2.4: Calculate the road segment queue Q based on the cost evaluation function C w The cost of each candidate segment in the robot segment set W is added to the candidate segment with the lowest cost, and the neighbor node x corresponding to the candidate segment is added to the robot segment set W. neighbor Add to the robot path node set N; and clear the path node queue Q N and segment queue Q w ; S2.5: Repeat S2.1 to S2.4 until the robot reaches the target state node x end As a neighbor node x neighbor The robot is added to the path node set N, and the initial planning path is obtained by combining the robot road segment set W and the robot path node set N; The S3 is specifically: S3.1: Construct a measurement matrix M based on the initial planned path. After performing singular value decomposition and calculation on the measurement matrix M, the super ellipsoid rotation matrix C is obtained. The formula is as follows: U∑V T ≡M C=Udiag{1,...,det(U)det(V)}V T Among them, a1 represents the starting point x of the current planned path start To the end point x of the initial planned path end direction, U,∑,V T are the left singular vector matrix, diagonal matrix and right singular vector transposed matrix respectively, || ||2 represents the Euclidean norm; I1 is the 1*n dimensional identity matrix, T represents the transpose, det() represents the determinant of the matrix; diag{} represents the diagonal matrix; S3.2: Based on the length of the current planned path and the starting point x start and the end point x end To establish a super ellipsoid, the formulas for related geometric parameters are as follows: Among them, a, b, and c represent the major axis, minor axis, and focal length of the super ellipsoid, respectively, and c best Represents the current planned path length, c min Represents the starting point x of the current planned path start and the end point x end the distance between them; S3.3: Construct the shape matrix L of the superellipsoid, as follows: S3.4: Randomly sample a set of nodes x within the unit hypersphere ball , the node x is rotated by the shape matrix L and the super ellipsoid rotation matrix C ball After scaling, rotating and translating, a sampling node group A within the super ellipsoid is obtained; S3.5: At each current path point x way Generate a corresponding set of random sampling nodes B in the neighborhood K of , and then connect the sampling node group A with each point x way After the sampling node group B is fused, the optimized sampling node set S is obtained; S3.6: Based on the robot arm target state node x end and the robot's initial state node x start , combined with the current sampling node optimization set S, generate the optimized planning path; S3.7: Repeat S3.1 to S3.6 until the optimal planned path is obtained. The S3.6 is specifically: S3.6.1: Initialize the robot path node set N′ and the robot segment set W′, and set the robot arm starting state node x start Put it into the robot path node set N′; S3.6.2: Randomly select a set of nodes from the sampling node optimization set S as potential path nodes and add them to the path node queue Q′ N In this group of nodes, the target state node x of the robot arm must be included. end and the robot's initial state node x start ; S3.6.3: For the path node queue Q′ N Each potential path node x′ in extend , establish the current potential path node x′ extend and each neighbor node x′ in its neighborhood D neighbor The potential connections between them, traversing the path node queue Q′ N Each potential path node x′ extend , thus obtaining the path node queue Q′ N All potential connections in S3.6.4: Use the three-dimensional model of the welding workpiece to calculate the path node queue Q′ N All potential connections in the collision detection are performed separately, and the potential connections that pass the collision detection are regarded as candidate links and added to the link queue Q′ w middle; S3.6.5: Calculate the road segment queue Q′ based on the cost evaluation function C w The cost of each candidate segment in the robot is added to the robot segment set W′ and the neighbor node x′ corresponding to the candidate segment is calculated. neighbor Add to the robot path node set N' and clear the path node queue Q' N and the road segment queue Q′ w ; S3.6.6: Repeat S3.6.1 to S3.6.5 until the robot reaches the target state node x end As a neighbor node x′ neighbor The robot path node set N′ is added, and the optimized planning path is obtained by combining the robot road segment set W′ and the robot path node set N′; Step 6: Send the welding path parameters to the robot control cabinet, and then control the welding robot to weld the workpiece.

2. A robot flexible welding path planning method based on dynamic batch sampling optimization according to claim 1, characterized in that: In the second step, a meta-heuristic algorithm is used to determine the welding sequence of the welds according to the welds of the welded workpiece.

3. The robot flexible welding path planning method based on dynamic batch sampling optimization according to claim 1 is characterized in that: In step 5, the welding path includes the planned path between the start and end points of the weld and the start or stop motion state of the robot and the planned path between two separated welds.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

6. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images