Workpiece adaptive welding method, medium, and apparatus

Abstract

The application provides a workpiece adaptive welding method, medium and equipment. The condition number of the Jacobian submatrix of the mechanical arm under each predicted configuration is calculated, the minimum collision distance between the robot prediction envelope and the fixture point cloud is obtained, whether the condition number of the Jacobian submatrix of the mechanical arm is less than the condition number threshold value, and whether the minimum collision distance is less than the set safety threshold value are judged, if neither is less than, the angular increment of each joint of the mechanical arm is solved to correct the corresponding mechanical arm joint vector, and the current welding seam position is welded; if any is less than, the mechanical arm external shaft speed and the mechanical arm joint speed are obtained to correct the mechanical arm external shaft vector and the mechanical arm joint vector, and the current welding seam position is welded. The real-time collaborative control of the visual sensor, the robot and the positioner cannot be realized in the prior art, so as to effectively cope with the thermal deformation in the welding process, improve the welding quality and production efficiency and solve other problems.

Description

Technical Field

[0001] This invention relates to the field of automated welding technology, and in particular to a workpiece adaptive welding method, medium, and equipment. Background Technology

[0002] Taking circumferential seam workpieces as an example, traditional automatic circumferential seam welding technology mainly uses dedicated automatic circumferential seam welding machines. These machines typically employ a vertical or horizontal structure and are suitable for welding cylindrical bodies with diameters ranging from 20 to 1000 mm. Their basic working principle is as follows: the cylindrical workpiece is fixed on the main shaft, which drives the workpiece to rotate via a continuously variable speed spindle. The welding torch maintains a fixed position or moves along a predetermined trajectory, thereby completing the circumferential seam welding. The main features of traditional automatic circumferential seam welding machines include: a pneumatic or hydraulic tailstock mechanism for workpiece clamping; an adjustable welding torch angle of 45 degrees for precise centering; and support for various welding processes such as TIG welding, MIG welding, and PLASMA welding.

[0003] However, this traditional approach has significant limitations. First, the welding path relies entirely on a pre-set program, making it unsuitable for workpiece assembly errors and thermal deformation during welding. Second, the equipment has poor adaptability; different workpiece sizes require readjustment of equipment parameters, and sometimes even replacement of specialized fixtures. Third, welding quality inspection can only be performed after welding is complete, making real-time monitoring and quality control during the process impossible.

[0004] To address the limitations of traditional methods in adapting to workpiece errors, vision sensor-based weld seam tracking technology has been developed in recent years. This technology integrates vision sensors into the welding system to identify the weld seam position in real time and adjust the welding trajectory based on the identification results. Currently, more advanced solutions employ 3D laser weld seam tracking sensors, which can identify weld seam position deviations in real time and automatically adjust the welding torch position through a control system to achieve real-time weld seam tracking. This technology can handle workpiece assembly errors and a certain degree of thermal deformation, with a dynamic compensation range typically within ±5mm.

[0005] However, existing visual tracking technologies still have shortcomings. Most systems employ pre-weld scanning and positioning, where a visual sensor scans the weld seam before welding begins, generates a welding path, and then welds according to that path. While this open-loop control method can compensate for assembly errors, it cannot handle the thermal deformation that occurs during welding. A few systems, although achieving real-time tracking during welding, are mainly applied to straight weld seams or simple curved weld seams. Their effectiveness is limited for circumferential welding of cylindrical workpieces, especially in scenarios requiring continuous workpiece rotation.

[0006] A comprehensive analysis of existing circumferential welding technologies for cylindrical workpieces reveals the following main problems: First, the thermal deformation compensation capability during the welding process is insufficient. Cylindrical workpieces undergo significant thermal deformation during circumferential welding, causing the weld position to shift. Traditional pre-programmed methods cannot cope with this dynamic change, and while existing weld tracking systems can provide some compensation, their response speed and compensation accuracy still need improvement.

[0007] Second, the collaborative control capability between the robot and the positioner is limited. Existing robot-positioner linkage systems mainly rely on pre-planned trajectories and lack a dynamic coordination mechanism based on real-time sensor information. When the weld position shifts, the system struggles to achieve coordinated adjustment between the robot and the positioner.

[0008] Third, the application of vision sensors is insufficient. In existing technologies, vision sensors are mainly used for pre-welding positioning or weld seam tracking at fixed positions, with limited real-time visual guidance that moves with the robot's end effector. This limits the effectiveness of vision sensors in complex welding scenarios.

[0009] Fourth, the system lacks flexibility and adaptability. When faced with cylindrical workpieces of different specifications and materials, the existing system often needs to be reprogrammed or its parameters adjusted, resulting in a long production preparation time and making it difficult to meet the needs of multi-variety, small-batch production.

[0010] Therefore, there is an urgent need for a new method for circumferential welding of cylindrical workpieces that can achieve real-time collaborative control of vision sensors, robots, and positioners, thereby effectively addressing thermal deformation during the welding process and improving welding quality and production efficiency. Summary of the Invention

[0011] Based on this, the purpose of this application is to provide an adaptive welding method, medium, and equipment for workpieces to solve at least one of the technical problems mentioned in the background art.

[0012] In a first aspect, this application provides a workpiece adaptive welding method, comprising: S1: Obtain the surface normal vector, weld direction tangent vector, and current weld position of each weld point on the workpiece surface within the scanning range of the acquisition device, so as to construct the absolute target pose matrix of the welding torch in the robot's global coordinate system and record the corresponding initial positioner external axis vector. S2: Obtain the global joint space vector of the current welding system, decouple to obtain the current external axis vector of the robotic arm, the joint vector of the robotic arm and the external axis vector of the positioner, calculate the external axis Jacobian submatrix and the robotic arm Jacobian submatrix respectively according to the current external axis position of the robotic arm and the angle of each joint of the robotic arm, and combine them to construct the global Jacobian matrix; S3: Obtain the welding path prediction sequence of several frames backward from the current time, transform the current weld position to the robot's global coordinate system, and correct the current weld position according to the initial external axis vector of the positioner and the current external axis vector of the positioner. Calculate the condition number of the robotic arm Jacobian submatrix under each prediction configuration, obtain the minimum collision distance between the robot's prediction envelope and the tooling fixture point cloud, and determine whether the condition number of the robotic arm Jacobian submatrix is ​​less than the condition number threshold and whether the minimum collision distance is less than the set safety threshold. S4: If none of them are less than, lock the current external axis vector of the robotic arm to solve the angular increment of each joint of the robotic arm, and correct the corresponding joint vector of the robotic arm according to the angular increment of each joint, weld the current weld position, and return to step S3. S5: If any is less than, unlock the current external axis vector of the robotic arm, calculate the Mohr-Ponrose pseudo-inverse of the global Jacobian matrix to construct the null projection operator of the global joint space; with the constraint that the welding torch tip does not deviate from the absolute trajectory pose velocity planned by vision, execute the set redundancy elimination equation to solve for the external axis velocity and joint velocity of the robotic arm, correct the external axis vector and joint vector of the robotic arm, weld the current weld position, and return to step S3.

[0013] Furthermore, the steps for constructing the absolute target pose matrix of the welding torch in the robot's global coordinate system include: Surface point clouds are collected, and voxel filtering and outlier removal are performed to obtain the point cloud of the weld area; Surface fitting is performed on the point cloud of the weld area to obtain the surface normal vector at the weld point, and B-spline curve fitting is performed on the weld trajectory points to obtain the weld direction tangent vector. A local orthogonal rotation matrix is ​​constructed based on the surface normal vector and the weld direction tangent vector. Combined with the preset process push angle and working angle, the absolute target pose matrix between the robot end effector and each weld point is constructed.

[0014] Furthermore, the steps of solving for the angular increments of each joint of the robotic arm and correcting the corresponding joint vectors of the robotic arm based on the angular increments of each joint include: Set the speed of the robot arm's external axis to zero to lock the current external axis of the robot arm; Construct a selection mask matrix to eliminate the partial derivative space of the robot arm's external axis in the global Jacobian matrix, and obtain the robot arm Jacobian submatrix; The Jacobian submatrix of the robotic arm is projected onto the low-dimensional active axis subspace to obtain the dimension-reduced active axis Jacobian matrix. The angular increments of each joint of the robotic arm are calculated based on the Jacobian matrix of the reduced active axis, and the corresponding joint vectors of the robotic arm are corrected based on the angular increments of each joint.

[0015] Further steps to obtain the external axis velocity and joint velocity of the robotic arm include: The TCP velocity vector is determined based on the absolute target pose matrix. The product of the Morpenrose pseudoinverse and the TCP velocity vector is obtained to get the tracking term velocity. Obtain the product matrix of the Morpenrose pseudoinverse and the Jacobian matrix, and derive the null space projection operator based on the difference between the identity matrix and the product matrix. The product of the null space projection operator and the optimal gradient vector is obtained to determine the null space optimization speed. Summing the tracking term velocity and the zero-space optimization velocity yields the global joint velocity, and decoupling the external axis velocity and the joint velocity of the robotic arm is then obtained.

[0016] Further steps to obtain the optimal gradient vector include: Obtain the obstacle repulsion distance index, construct a first objective function with the goal of maximizing the obstacle repulsion distance, and solve for the optimal gradient value of the obstacle repulsion distance index; Obtain the kinematic operability index, construct an objective function that maximizes the kinematic operability, and solve for the optimal gradient value of the kinematic operability index. Obtain the joint limit avoidance index, construct an objective function that maximizes the joint limit avoidance performance, and solve for the optimal gradient value of the joint limit avoidance index. Based on the set weighting coefficients, the optimal gradient values ​​of the obstacle rejection distance index, kinematic maneuverability index, and joint limit avoidance index are weighted and summed to obtain the optimal gradient vector.

[0017] Further, the step of obtaining the optimal gradient value of the obstacle repulsion distance index includes: Calculate the minimum Euclidean distance between the robot's link surface and the tooling fixture point cloud in the current configuration, and determine whether the minimum Euclidean distance is less than or equal to the set safety threshold. If so, then construct the obstacle repulsion distance index function based on the set potential field gain coefficient, minimum Euclidean distance and set safety threshold, and take the derivative of the obstacle repulsion distance index function with respect to each joint variable to obtain the optimal gradient value of the obstacle repulsion distance index. If not, the optimal gradient value of the obstacle repulsion distance index is the zero vector.

[0018] Further steps for obtaining the optimal gradient value of the kinematic operability index include: Calculate the determinant of the product matrix of the robot's global Jacobian matrix and its transpose, and take the square root to obtain the kinematic maneuverability index function. The optimal gradient value of the kinematic operability index is obtained by taking the partial derivatives of the kinematic operability index function with respect to each joint variable.

[0019] Further, the steps for obtaining the optimal gradient value of the joint constraint avoidance index include: Obtain the upper physical limit angle, lower physical limit angle, and current angle of each joint, and calculate the midpoint coordinates of the stroke of each joint to construct a penalty function for joint deviation from the center position, and take the negative value to obtain the joint limit avoidance index function. The optimal gradient value of the joint limit avoidance index is obtained by taking the partial derivative of the joint limit avoidance index function with respect to each joint variable.

[0020] Secondly, this application also provides a computer storage medium storing executable program code; the executable program code is used to execute the workpiece adaptive welding method described in any one of the first aspects.

[0021] Thirdly, this application also provides a terminal device, including a memory and a processor; the memory stores program code executable by the processor; the program code is used to execute the workpiece adaptive welding method according to any one of the first aspects.

[0022] This application provides a workpiece adaptive welding method, medium, and equipment. The method involves: S1: acquiring the surface normal vector, weld direction tangent vector, and current weld position at each weld point on the workpiece surface within the scanning range of the acquisition device to construct the absolute target pose matrix of the welding torch in the robot's global coordinate system, and recording the corresponding initial positioner external axis vector; S2: acquiring the current global joint space vector of the welding system, decoupling to obtain the current robot arm external axis vector, robot arm joint vector, and positioner external axis vector, and calculating the robot arm external axis Jacobian sub-matrix and robot arm Jacobian sub-matrix respectively based on the current robot arm external axis position and the angles of each robot arm joint, and combining them to construct the global Jacobian matrix; S3: acquiring a welding path prediction sequence of several frames forward from the current moment, and transforming the current weld position to the robot's global coordinate system based on the initial and current positioner external axis vectors to calculate the bar of the robot arm Jacobian sub-matrix under each prediction configuration. The number of parts is calculated, and the minimum collision distance between the robot's predicted envelope and the tooling fixture point cloud is obtained. It is determined whether the condition number of the robotic arm's Jacobian submatrix is ​​less than the condition number threshold and whether the minimum collision distance is less than the set safety threshold. S4: If neither is less than the threshold, the current robotic arm's external axis vector is locked to solve the angular increment of each joint of the robotic arm. The corresponding robotic arm joint vector is corrected according to the angular increment of each joint. Welding is performed on the current weld position, and the process returns to step S3. S5: If either is less than the threshold, the current robotic arm's external axis vector is unlocked. The Mohr'Ponrose pseudo-inverse of the global Jacobian matrix is ​​calculated to construct the null space projection operator of the global joint space. With the constraint that the welding torch tip does not deviate from the absolute trajectory pose velocity planned by vision, the set redundancy elimination equation is executed to solve for the robotic arm's external axis velocity and robotic arm joint velocity. The robotic arm's external axis vector and robotic arm joint vector are corrected. Welding is performed on the current weld position, and the process returns to step S3. It improves upon existing technologies that cannot achieve real-time collaborative control of vision sensors, robots, and positioners, thereby effectively addressing issues such as thermal deformation during welding and improving welding quality and production efficiency. Attached Figure Description

[0023] Figure 1 This is a flowchart of an embodiment of the adaptive welding method for workpieces of the present invention; Figure 2 This is an overall flowchart of a welding system according to an embodiment of the adaptive welding method for workpieces of the present invention. Figure 3 This is a flowchart illustrating a method for correcting the external axis vector and / or joint vector of a robotic arm, as an embodiment of the adaptive welding method for workpieces of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] It should be noted that if the embodiments of the present invention involve directional indications, such as up, down, left, right, front, back, etc., these directional indications are only used to explain the relative positional relationships and movement of the components in a specific posture. If the specific posture changes, the directional indications will also change accordingly. Furthermore, if the embodiments of the present invention involve descriptions such as "first," "second," "S1," "S2," "step one," "step two," etc., these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance, or implicitly indicating the number of technical features indicated or the execution order of the method. Those skilled in the art will understand that anything that does not violate the inventive concept should be included within the scope of protection of the present invention.

[0026] like Figure 1 As shown, the present invention provides a workpiece adaptive welding method, comprising: S1: Obtain the surface normal vector, weld direction tangent vector, and current weld position of each weld point on the workpiece surface within the scanning range of the acquisition device, so as to construct the absolute target pose matrix of the welding torch in the robot's global coordinate system and record the corresponding initial positioner external axis vector. Specifically, point cloud data or mesh models of the workpiece surface within the scanning range of the acquisition device can be obtained through methods such as 3D vision scanning and laser contour measurement. Then, through steps such as surface fitting, principal component analysis, or differential geometry calculation, local geometric features at each welding point are extracted to obtain the surface normal vector, weld direction tangent vector, and current weld position. This allows the construction of the absolute target pose matrix of the welding torch in the robot's global coordinate system, while simultaneously recording the corresponding initial external axis vector of the positioner, providing a data foundation for subsequent correction and coordinate transformation steps. The acquisition device can be a commonly used 3D vision sensor, structured light 3D scanner, line laser scanner, or binocular stereo vision system.

[0027] In a preferred embodiment, the present invention provides a welding system hardware consisting of a main control computer, a multi-axis positioner, a three-dimensional vision sensor, and a highly redundant robot actuator (including a gantry and a multi-joint robotic arm). The overall flowchart of the welding system is shown below. Figure 2 As shown, it includes: After the system starts, the main control computer starts parallel vision recognition threads and motion planning threads, which communicate asynchronously through a first-in-first-out (FIFO) shared data buffer queue. Synchronous sampling and buffering: The vision sensor scans the weld seam at a constant time step at high frequency, capturing the local coordinate data of the current welding point in real time. Record the joint angle of the positioner at this time. Pack it and write it into the data cache queue.

[0028] Delayed dequeueing and restoration: Motion planning threads based on process welding speed Fixed physical distance between sensor and welding torch Calculate the precise delay time .go through After a certain time, the algorithm reads historical data from the queue and combines it with the current instantaneous positioner angle. and external axis vector of the positioner The scan coordinates are accurately restored to the current actual welding target position. : ;

[0029] In a preferred embodiment, the step of constructing the absolute target pose matrix of the welding torch in the robot's global coordinate system includes: S11: Collect surface point clouds, perform voxel filtering and outlier removal to obtain point clouds of the weld area; S12: Perform surface fitting on the point cloud of the weld area to obtain the surface normal vector at the weld point, and perform B-spline curve fitting on the weld trajectory points to obtain the weld direction tangent vector. Specifically, a motion planning system can be used to perform voxel filtering and outlier removal on the point cloud of the region of interest (ROI) captured by the 3D vision sensor. Subsequently, principal component analysis (PCA) and singular value decomposition (SVD) are used to fit the local surface, and the surface normal vector at the current solder point is calculated in real time. Simultaneously, the weld trajectory points are fitted with B-spline curves and differentiated to obtain the weld direction tangent vector. .

[0030] S13: Construct a local orthogonal rotation matrix based on the surface normal vector and the weld direction tangent vector, and combine it with the preset process push angle and working angle to construct the absolute target pose matrix between the robot end effector and each weld point.

[0031] Specifically, the process advance angle (also known as the travel angle) is the angle between the welding torch axis and the weld normal (or a plane perpendicular to the weld direction) in the welding advance direction; the working angle (also known as the working angle) is the angle between the welding torch axis and the workpiece surface normal (or the weld tangent plane), describing the tilting posture of the welding torch in a plane perpendicular to the welding direction, and can be obtained by cross product to obtain the secondary normal vector. Construct a local orthogonal rotation matrix and combine it with a preset process angle. and working angle Generate the final absolute target pose matrix that is sent to the robot for execution. : ;

[0032] Therefore, the welding torch can dynamically "adaptively deflect" according to the local ellipticity distortion or twist of the workpiece where the welding point is located, and always maintain an absolutely precise process angle with the workpiece surface, that is, the angle between the welding torch and the plane where the welding point is located when the welding torch is welding each welding point.

[0033] S2: Obtain the global joint space vector of the current welding system, decouple to obtain the current external axis vector of the robotic arm, the joint vector of the robotic arm and the external axis vector of the positioner, calculate the external axis Jacobian submatrix and the robotic arm Jacobian submatrix respectively according to the current external axis position of the robotic arm and the angle of each joint of the robotic arm, and combine them to construct the global Jacobian matrix; Specifically, this involves a large-stroke gantry crane (whose translational external axis can be generalized as an m-dimensional linear axis system, such as...) This corresponds to linear prism joint groups such as J1, J1+J2, and J1+J2+J3, respectively, working in collaboration with a 6-axis industrial multi-joint robotic arm at the end and a k-axis positioner, achieving extremely high redundancy (total degrees of freedom). Taking a welding system as an example, the global joint space state vector of the current welding system can be optionally defined as follows: It is decoupled and divided into three sub-state spaces: ;

[0034] in, For gantry External axis vector of a linear robotic arm; Let be the joint vector of the end-effector 6-axis robot arm with a kinematically closed solution (analytical solution); This is the external axis vector of the positioner.

[0035] Let the forward kinematic equation of the end-effector (TCP) of the system be: Then the global Jacobian matrix in the global system state Defined as: ;

[0036] In the formula, Let Jacobian submatrix be the external axis of the robotic arm. For the 6-axis robotic arm, the robotic arm Jacobian submatrix is ​​used. Since the external axis vector of the positioner obtained by decoupling is used for the transformation of the weld position in subsequent steps, the pose of the positioner has been taken into account. Therefore, this step only needs to use the external axis Jacobian submatrix of the robotic arm and the robotic arm Jacobian submatrix to construct the global Jacobian matrix.

[0037] S3: Obtain the welding path prediction sequence of several frames backward from the current time, transform the current weld position to the robot's global coordinate system, and correct the current weld position according to the initial external axis vector of the positioner and the current external axis vector of the positioner. Calculate the condition number of the robotic arm Jacobian submatrix under each prediction configuration, obtain the minimum collision distance between the robot's prediction envelope and the tooling fixture point cloud, and determine whether the condition number of the robotic arm Jacobian submatrix is ​​less than the condition number threshold and whether the minimum collision distance is less than the set safety threshold. Specifically, because traditional welding control lacks the ability to proactively assess subsequent paths, it often only responds passively after a problem occurs, leading to welding interruptions or quality defects. A possible solution is to retrieve the current time step forward from the cache queue. The system predicts the frame sequence and then transforms the current weld position to the robot's global coordinate system. Since the positioner's pose may change during welding, the transformed weld position can be corrected using the initial and current external axis vectors of the positioner. Finally, the Jacobian matrix of the robot's predicted configuration is calculated in real-time. The condition number is used to determine whether the condition number of the robotic arm's Jacobian submatrix is ​​less than a condition number threshold to assess whether it is about to fall into a kinematic singularity; simultaneously, the minimum distance between the robot's predicted envelope and the visually captured tooling fixture point cloud is calculated. It determines whether the minimum collision distance is less than a set safety threshold to assess whether there is a collision risk, thus constructing a dual-threshold online assessment system of "operability + collision distance". This enables the control system to predict the risk of kinematic degradation and collision hazards in advance, providing a scientific basis for switching hierarchical control strategies. It realizes the transformation of the control paradigm from passive protection to active prevention, and significantly improves the continuous operation capability and safety under complex working conditions.

[0038] S4: If none of them are less than, lock the current external axis vector of the robotic arm to solve the angular increment of each joint of the robotic arm, and correct the corresponding joint vector of the robotic arm according to the angular increment of each joint, weld the current weld position, and return to step S3. Specifically, such as Figure 3As shown, when both threshold criteria are met, i.e. the condition number of the robotic arm's Jacobian submatrix is ​​not less than the condition number threshold and the minimum collision distance is not less than the set safety threshold, and the welding system is on a normal welding path without singularities or collision risks, it indicates that the current configuration and the look-ahead path are in a safe area with good operability and no collision risks. At this time, locking the external axis of the robotic arm and completing trajectory tracking only through the joint space of the robotic arm can significantly reduce the control dimension and computational overhead, while avoiding mechanical wear and vibration transmission caused by frequent micro-movements of the external axis. Under the premise of ensuring trajectory accuracy, the optimal allocation of control resources and the efficient and stable operation of the mechanical system can be achieved.

[0039] In a preferred embodiment, the steps of solving for the angular increments of each joint of the robotic arm and correcting the corresponding joint vectors of the robotic arm based on the angular increments of each joint include: S41: Set the speed of the robot arm's external axis to zero to lock the current robot arm's external axis; Specifically, the speed of the m-dimensional linear external axis of the gantry can be optionally forced to be... That is, within the current high-frequency fine-tuning cycle of vision, it is treated as a known constant: ;

[0040] S42: Construct a selection mask matrix to eliminate the partial derivative space of the robot arm's external axis in the global Jacobian matrix, and obtain the robot arm Jacobian submatrix; Specifically, an optional active axis selection mask matrix can be constructed: the main control computer internally constructs a matrix with a dimension of... Dynamic active axis selection mask matrix Completely eliminate the partial derivative space of the outer axis of the gantry: ;

[0041] S43: Project the Jacobian submatrix of the robotic arm onto the low-dimensional active axis subspace to obtain the dimension-reduced active axis Jacobian matrix; Specifically, the global Jacobian matrix can be projected onto a low-dimensional active axis subspace, at which point the reduced active axis Jacobian matrix... Perfect collapse into phalanx: ;

[0042] S44: Solve for the angular increment of each joint of the robotic arm based on the reduced-dimensional Jacobian matrix of the active axis, and correct the corresponding joint vector of the robotic arm based on the angular increment of each joint.

[0043] Specifically, since the external axes are completely removed, the solution space is reduced to only the standard 6-axis industrial robot that satisfies the Pieper criterion (the three end-effectors intersect at a single point). The system directly stops using time-consuming numerical iterations (such as Newton-Raphson) and instead directly calls the underlying robotic arm's geometrically closed-form analytical solution to obtain the angular increments of each joint of the robotic arm: ;

[0044] This branch completely eliminates the need for heavy gantry cranes during routine position tracking, enabling inverse kinematics calculations to leap from milliseconds to microseconds (<1ms), fundamentally eliminating severe joint jitter and disordered jumps in high-frequency visual servoing of ultra-redundant systems.

[0045] S5: If any is less than, unlock the current external axis vector of the robotic arm, calculate the Mohr-Ponrose pseudo-inverse of the global Jacobian matrix to construct the null projection operator of the global joint space; with the constraint that the welding torch tip does not deviate from the absolute trajectory pose velocity planned by vision, execute the set redundancy elimination equation to solve for the external axis velocity and joint velocity of the robotic arm, correct the external axis vector and joint vector of the robotic arm, weld the current weld position, and return to step S3.

[0046] Specifically, such as Figure 3 As shown, when the condition number of the Jacobian submatrix of the robotic arm is less than the condition number threshold, or the minimum collision distance is less than the set safety threshold, indicating insufficient operability or collision risk, the external axis lock can be released and global redundancy resolution can be initiated. A zero-space projection operator is constructed through the Moore-Penrose pseudo-inverse, with the primary constraint of strictly ensuring the accuracy of welding torch tip trajectory tracking. Multi-objective performance indicators are optimized in the zero space, enabling the system to make full use of all redundant degrees of freedom for adaptive adjustment without sacrificing welding trajectory accuracy when facing kinematic constraints or environmental obstacles. This achieves decoupled control of accuracy maintenance and performance improvement, significantly expanding the achievable welding range of the robot in narrow spaces or complex postures.

[0047] In a preferred embodiment, the steps of determining the external axis velocity and joint velocity of the robotic arm include: S51: Determine the TCP velocity vector based on the absolute target pose matrix, obtain the product of the Morpenrose pseudoinverse and the TCP velocity vector, and get the tracking term velocity; S52: Obtain the product matrix of the Morpenrose pseudoinverse and the Jacobian matrix, and obtain the null space projection operator based on the difference between the identity matrix and the product matrix. S53: Obtain the product of the null space projection operator and the optimal gradient vector to get the null space optimization speed; S54: Summing the tracking term velocity and the zero-space optimization velocity, we obtain the global joint velocity, and decouple the external axis velocity and the joint velocity of the robotic arm.

[0048] Specifically, since the absolute target pose matrix describes the target pose of the welding torch at a single welding point, and welding is a continuous process that requires connecting discrete target points into a continuous pose trajectory, the TCP velocity vector can be determined in real time based on the current pose of the robotic arm and the next target pose. Furthermore, since traditional redundancy resolution methods often use minimum norm solutions or simple projections, it is difficult to achieve precise decoupling between the main task tracking and redundancy optimization. Therefore, the tracking velocity can be obtained by multiplying the Moore-Penrose pseudoinverse with the TCP velocity vector, ensuring the precise execution of the main task (welding torch tip trajectory tracking). Then, the Moore-Penrose pseudoinverse and the Jacobian... The product matrix of matrices constructs the null space projection operator, which projects the optimal gradient vector onto the Jacobian null space to obtain the null space optimized velocity. Finally, the velocity of the tracking term is summed with the null space optimized velocity to obtain the global joint velocity and decouple them. Through the rigorous mathematical framework of "pseudo-inverse tracking + null space projection + velocity superposition", the tracking error of the main task is theoretically unaffected by the null space motion, realizing the complete decoupling between the task space and the null space motion. At the same time, through the optimized configuration of the null space velocity, the redundant degrees of freedom can be fully utilized without interfering with the welding trajectory accuracy, providing a clear mathematical framework and a reliable numerical implementation path for subsequent multi-objective performance optimization.

[0049] In a preferred embodiment, once the forward prediction sliding window triggers a singularity warning or a tooling interference warning ( or The system immediately releases the static lock on the external axis of the gantry and opens the global lock. Dimensional redundancy space: (1) Unlock and restore the global redundant Jacobian: Reactivate the Lungmen sub-Jacobi matrix The system's global generalized velocity vector is recovered as .

[0050] (2) Solving the null projection matrix: Calculate the Moore-Ponrose pseudoinverse of the global non-square Jacobian matrix. And construct the null projection operator for the global joint space. : ;

[0051] (3) Solving the full-space macro-micro redundancy avoidance control equations: under strict guarantee, the welding torch tip (TCP) moves along the absolute trajectory planned by vision with the pose and velocity of the torch tip (TCP) in the vision. Under hard constraints that do not cause any deviations, perform the following redundant resolution equations: ;

[0052] in, For global joint velocity, For the external axis speed of the robotic arm, For the joint speed of the robotic arm, The Morpenrose pseudoinverse of the global Jacobian matrix. For TCP velocity vectors, It is the identity matrix. The global Jacobian matrix. This is the optimal gradient vector.

[0053] (4) Cooperative avoidance execution: Based on the orthogonal properties of the projection operator, additional joint velocities are added. The pose response mapped to the TCP end is strictly zero. Based on this, the system drives the entire heavy-duty system in tandem: utilizing the external axis of the long-stroke gantry. Perform overall macroscopic translation and sliding, in conjunction with a 6-axis robotic arm. The in-situ microscopic attitude reversal, in conjunction with the action, brings the entire linkage system away from the singular region and the fixture interference region.

[0054] In a preferred embodiment, the step of obtaining the optimal gradient vector includes: S531: Obtain the obstacle repulsion distance index, construct the first objective function with the goal of maximizing the obstacle repulsion distance, and solve for the optimal gradient value of the obstacle repulsion distance index; S532: Obtain the kinematic operability index, construct an objective function that maximizes the kinematic operability, and solve for the optimal gradient value of the kinematic operability index; S533: Obtain the joint limit avoidance index, construct an objective function with the goal of maximizing the joint limit avoidance performance, and solve for the optimal gradient value of the joint limit avoidance index. S534: Based on the set weight coefficients, the optimal gradient values ​​of the obstacle rejection distance index, kinematic maneuverability index, and joint limit avoidance index are weighted and summed to obtain the optimal gradient vector.

[0055] Specifically, to achieve the coordinated optimization of multiple objectives such as obstacle repulsion distance, kinematic maneuverability, and joint constraint avoidance, the system constructs the objective function corresponding to the optimal gradient vector as a linear weighted combination of the obstacle repulsion distance, kinematic maneuverability, and joint constraint avoidance indices: ;

[0056] in, , , These are the weighting coefficients for the obstacle rejection distance index, kinematic maneuverability index, and joint constraint avoidance index, respectively, and the optimal gradient vector. That is, the scalar function applied to each joint variable The vector of partial derivatives: ;

[0057] In a preferred embodiment, the step of obtaining the optimal gradient value of the obstacle repulsion distance index includes: S5311: Calculate the minimum Euclidean distance between the robot's link surface and the tooling fixture point cloud in the current configuration, and determine whether the minimum Euclidean distance is less than or equal to the set safety threshold. S5312: If so, construct the obstacle repulsion distance index function based on the set potential field gain coefficient, minimum Euclidean distance and set safety threshold, and take the derivative of the obstacle repulsion distance index function with respect to each joint variable to obtain the optimal gradient value of the obstacle repulsion distance index. S5313: If not, the optimal gradient value of the obstacle repulsion distance index is the zero vector.

[0058] Specifically, the obstacle repulsion distance index is constructed based on the Artificial Potential Field method to ensure that any link of the robot stays away from spatial obstacles such as tooling fixtures during movement.

[0059] (a) Distance calculation: Suppose the robot is in its current configuration Below, the minimum Euclidean distance between the link surface and the obstacle point cloud is: The set absolute safety distance threshold is .

[0060] (b) Index Function: To maximize safety performance, the obstacle repulsion distance index is defined as the negative value of the repulsion potential energy. When the robot approaches the obstacle, the potential energy increases sharply, and the index function value decreases sharply. ;

[0061] in, is the potential field gain coefficient.

[0062] (c) Gradient calculation Differentiating the joint variables using the chain rule: ;

[0063] In the formula, Let this be the local Jacobian matrix of the point on the robot closest to the obstacle. This is the unit direction vector pointing from the nearest point on the machine to the nearest point on the obstacle. This gradient direction generates a "virtual repulsive force" that is directly mapped onto the joint torque, forcing the corresponding joint to perform an avoidance torsion.

[0064] In a preferred embodiment, the step of obtaining the optimal gradient value of the kinematic operability index includes: S5321: Calculate the determinant of the product matrix of the robot's global Jacobian matrix and its transpose, and take the square root to obtain the kinematic maneuverability index function. S5322: Calculate the partial derivatives of the kinematic maneuverability index function with respect to each joint variable to obtain the optimal gradient value of the kinematic maneuverability index.

[0065] Specifically, the kinematic manipulation index is used to assess how far the current robot configuration is from the kinematic singularity, and the Yoshikawa manipulation measure is used.

[0066] (a) Index function: defined as the robot's global Jacobian matrix The volume of the operational ellipsoid formed: ;

[0067] As the robotic arm approaches the singularity, the Jacobian matrix decreases in rank. It approaches 0; the larger the value, the higher the flexibility of the robotic arm.

[0068] (b) Gradient calculation The gradient of this term is intended to guide the robotic arm to move in the direction of increasing volume of the manipulated ellipsoid.

[0069] In the actual system's underlying solution, this partial derivative is quickly obtained by calculating the condition number of the Jacobian matrix or the numerical solution of the determinant difference corresponding to the small increments of each joint.

[0070] In a preferred embodiment, the step of obtaining the optimal gradient value of the joint constraint avoidance index includes: S5331: Obtain the upper physical limit angle, lower physical limit angle and current angle of each joint, and calculate the midpoint coordinates of the stroke of each joint to construct a penalty function for joint deviation from the center position, and take the negative value to obtain the joint limit avoidance index function. S5332: Calculate the partial derivatives of the joint limit avoidance index function with respect to each joint variable to obtain the optimal gradient value of the joint limit avoidance index.

[0071] Specifically, since a joint is very likely to reach its physical rotation limit during continuous large-angle circumferential welding, the joint limit avoidance index is used to keep all joints as close as possible to the center of their physical stroke.

[0072] (a) Index function: defined as the negative value of the penalty function for joint deviation from the center position: ;

[0073] in, This represents the total number of joints in the robotic arm. and For the first The upper and lower physical limits of each joint. For the first The coordinates of the midpoint of the stroke of each joint, i.e. .

[0074] (b) Gradient calculation : Regarding the first The partial derivatives of each joint are: ;

[0075] This gradient vector provides a "reset spring force," causing joints approaching their physical limits to tend to retract towards the midpoint.

[0076] In this embodiment, an adaptive welding method for workpieces according to the present invention is presented, the core inventive point of which is: I. A five-stage closed-loop control framework of "visual perception - global modeling - prediction and evaluation - hierarchical control - zero-space optimization" is introduced. Compared with traditional circumferential welding methods based on teaching and reproduction or fixed trajectories, this framework offers unique technical advantages such as high trajectory accuracy, strong dynamic adaptability, large safety redundancy, and superior multi-objective collaboration. Specifically, these advantages include: 1. A welding torch absolute pose construction mechanism based on point cloud surface fitting and B-spline interpolation enables precise perception and process adaptation of complex curved surface welds: Traditional circumferential weld methods often rely on manual teaching or offline programming based on CAD models, which is difficult to adapt to weld position drift caused by actual workpiece machining errors, clamping deviations, and thermal deformation, and lacks a coordinated consideration of local surface geometry and welding process parameters (push angle, working angle). This application acquires the workpiece surface point cloud through acquisition equipment, obtains the weld area point cloud through voxel filtering and outlier removal, then extracts the surface normal vector at the weld point through surface fitting, and obtains the weld direction tangent vector through B-spline curve fitting. After constructing a local orthogonal rotation matrix, combined with preset process push angle and working angle, the absolute target pose matrix of the welding torch in the robot's global coordinate system is generated. This collaborative mechanism of "point cloud filtering-surface reconstruction-process embedding" enables the welding trajectory to accurately fit the actual workpiece geometry, effectively compensate for machining and clamping errors, and ensure that the welding torch posture always meets the process specification requirements, providing a high-precision target input reference for subsequent closed-loop control.

[0077] 2. A global joint space decoupling and hierarchical Jacobian modeling mechanism to achieve a unified kinematic representation of the multi-axis system of robotic arm-external axis-positioner: Traditional welding robot control often treats the robotic arm and external axes (walking axis, positioner) separately, lacking a globally coordinated kinematic model, leading to trajectory tracking errors or dynamic conflicts during multi-axis linkage. This application obtains the global joint space vector of the current welding system, decouples the robotic arm external axis vector, robotic arm joint vector, and positioner external axis vector, and calculates the robotic arm external axis Jacobian submatrix and robotic arm Jacobian submatrix respectively, combining them to construct the global Jacobian matrix. This "decoupling-blocking-combination" modeling strategy enables the robotic arm, walking axis, and positioner to achieve kinematic coupling within a unified mathematical framework, laying a rigorous differential kinematic foundation for subsequent global redundancy resolution and multi-objective optimization, effectively avoiding trajectory distortion and synchronization failure problems caused by separate axis control.

[0078] 3. An adaptive hierarchical control mechanism based on look-ahead path prediction and dual-threshold criteria enables online dynamic evaluation and mode switching of operability and collision safety: Traditional welding methods lack the ability to anticipate subsequent paths, often triggering protective shutdown only when joint limits or collision risks have actually occurred, leading to welding interruptions and quality defects. This application obtains several frames of welding path prediction sequences forward from the current moment, calculates the condition number (operability index) of the robotic arm Jacobian submatrix under each prediction configuration, and simultaneously obtains the minimum collision distance between the robot's prediction envelope and the tooling fixture point cloud, constructing a dual-threshold criterion system of "condition number threshold + safety threshold". When both indices meet the requirements, the external axis of the robotic arm is locked, and trajectory tracking is completed only through the joint space of the robotic arm; when either index exceeds the limit, the external axis is unlocked, and global redundancy resolution is initiated. This adaptive mechanism of "prediction evaluation - threshold determination - hierarchical switching" enables the control system to perceive potential kinematic degradation and collision risks in advance, maximizing the use of system redundancy resources while ensuring safety, and significantly improving continuous welding capability and dynamic adaptability under complex working conditions.

[0079] 4. A global redundancy resolution mechanism based on null-space projection and multi-objective gradient optimization achieves accurate trajectory tracking and synergistic improvement of multiple performance indicators: Traditional redundant robot control often adopts a single optimization objective (such as the minimum joint velocity norm), which is difficult to simultaneously consider multiple constraints such as obstacle avoidance, maneuverability maintenance, and joint limit avoidance. In this application, under the mode of unlocking the external axis, the Moore-Penrose pseudo-inverse of the global Jacobian matrix is ​​calculated to construct the null-space projection operator of the global joint space; with the strict constraint that the welding torch tip does not deviate from the absolute trajectory pose velocity planned by vision, the redundancy resolution equation is executed, and the tracking velocity is superimposed with the null-space optimized velocity to obtain the external axis velocity and joint velocity of the robot arm. The null-space optimized velocity is composed of the weighted sum of the optimal gradient values ​​of the obstacle repulsion distance index, the kinematic maneuverability index, and the joint limit avoidance index. This hierarchical redundancy resolution architecture, which combines "precise tracking of the main task with multi-objective optimization in zero space," theoretically ensures that trajectory tracking accuracy is unaffected by zero-space motion. At the same time, it maximizes obstacle repulsion distance, improves kinematic maneuverability, and avoids joint limitations by utilizing redundant degrees of freedom, thus achieving multi-objective collaborative optimization of welding quality, safety, and mechanical life.

[0080] II. A multi-objective gradient fusion strategy based on artificial potential field, operability metric, and joint center deviation penalty is adopted. Compared with single-objective or fixed-weight optimization methods, this strategy exhibits unique technical advantages, including more sensitive obstacle avoidance response, stable operability maintenance, and reasonable joint distribution. Specifically, this is reflected in the following aspects: 1. An obstacle repulsion mechanism based on distance threshold triggering and potential field gradient calculation enables real-time collision risk perception and proactive avoidance: Traditional obstacle avoidance methods often rely on global path planning or passive collision detection, resulting in high computational overhead and slow response. This application calculates the minimum Euclidean distance between the link surface and the tooling fixture point cloud under the current robot configuration and compares it with a set safety threshold. Only when the minimum distance is less than or equal to the safety threshold is an obstacle avoidance performance index function constructed based on the potential field gain coefficient, the minimum distance, and the safety threshold, and the repulsion gradient is generated by differentiation; otherwise, the gradient is a zero vector. This sparse activation mechanism of "threshold triggering - local potential field - conditional calculation" ensures that obstacle avoidance calculation is triggered only when near the danger zone, significantly reducing the computational burden under normal working conditions. At the same time, by guiding joint movement away from obstacles through the potential field gradient, proactive, smooth, and predictable collision risk avoidance is achieved, avoiding the adverse effects of sudden shutdowns or emergency braking on weld quality.

[0081] 2. A maneuverability optimization mechanism based on the condition number determinant of the Jacobian matrix enables online maintenance of the robotic arm's kinematic performance: Traditional methods often neglect the maneuverability degradation caused by changes in the robotic arm's posture during welding, which can easily lead the robot to approach singular configurations, causing speed loss and amplified tracking errors. This application uses the square root of the determinant of the product of the global Jacobian matrix and its transpose as the kinematic maneuverability index function, and obtains the maneuverability gradient by taking its partial derivatives with respect to each joint variable. This "determinant measurement-analytical differentiation-gradient ascent" maneuverability optimization strategy enables the control system to actively drive the robotic arm to move away from singular configurations, maintaining the uniform distribution and controllability of the velocity of each joint, effectively suppressing the decline in trajectory tracking accuracy and vibration problems caused by insufficient maneuverability, and ensuring the dynamic response quality during high-speed welding.

[0082] 3. A limit avoidance mechanism based on joint midpoint penalty to achieve balanced utilization of joint workspace and extended mechanical life: Traditional welding trajectory planning often leaves certain joints near their extreme positions for extended periods, accelerating mechanical wear and increasing control difficulty. This application obtains the physical upper limit angle, physical lower limit angle, and current angle of each joint, calculates the midpoint coordinates of the stroke, constructs a penalty function for joint deviation from the center position, takes the negative value to obtain the limit avoidance index function, and then derives the limit avoidance gradient. This "center deviation penalty - gradient descent - workspace centering" limit avoidance strategy makes each joint tend to be distributed near the midpoint of the stroke during welding, avoiding kinematic degradation and control saturation caused by one or more joints approaching their extreme positions, effectively extending the service life of the joint servo system, while maintaining the robot's flexible response capability throughout the entire workspace.

[0083] The inventive points of this invention also include: A six-DOF pose-adaptive visual guidance method and architecture for circumferential seams Protecting an asynchronous queue and delay time based on dual threads (visual recognition thread and motion planning thread) The control method of the compensation architecture is characterized by extracting local point clouds using a three-dimensional vision sensor and calculating surface normal vectors. tangent vector A local orthogonal affine coordinate system is constructed and a preset process push angle and working angle are integrated to generate an absolute target attitude matrix in real time to compensate for thermal deformation and distortion, thereby achieving six-degree-of-freedom joint closed-loop correction of position and attitude.

[0084] A method for dynamic dimensionality reduction and inverse kinematics solution for highly redundant multi-axis systems In the high-frequency visual tracking process of a 6-axis robotic arm system containing m linear external axes, protection is achieved by constructing a dynamic active axis selection mask matrix to forcibly zero (lock) the velocities of the m-dimensional external axes; and by collapsing and reducing the dimensionality of the non-square global Jacobian matrix to... The matrix, and thus bypassing time-consuming numerical iterations, directly calls the underlying kinematic control method of the robotic arm's inherent geometric closed analytical solution for extremely rapid convergence.

[0085] 3. A continuous welding adaptive avoidance method combining forward prediction and null space mapping: A method is provided to predict the Jacobian condition number (singularity) and interference envelope distance of a robotic arm M frames in advance using a sliding window; characterized in that, when a safety threshold is triggered, the pseudo-inverse of the Jacobian matrix is ​​utilized. and its null spatial projection operator Construct governing equations; track welding trajectory speed at the strictly constrained tool center point (TCP). Under the premise of constantness, the gradient of the objective cost function is projected onto the null space, driving redundant joints to generate additional actions to autonomously avoid singularities and obstacles.

[0086] On the other hand, the present invention also provides a computer storage medium storing executable program code; the executable program code is used to execute any of the above-described adaptive welding methods for workpieces.

[0087] On the other hand, the present invention also provides a terminal device, including a memory and a processor; the memory stores program code that can be executed by the processor; the program code is used to execute any of the above-mentioned adaptive welding methods for workpieces.

[0088] For example, the program code can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the program code in the terminal device.

[0089] The terminal device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the terminal device may also include input / output devices, network access devices, buses, etc.

[0090] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0091] The memory can be an internal storage unit of the terminal device, such as a hard drive or RAM. The memory can also be an external storage device of the terminal device, such as a plug-in hard drive, SmartMediaCard (SMC), Secure Digital (SD) card, or FlashCard. Furthermore, the memory can include both internal and external storage units of the terminal device. The memory is used to store the program code and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or will be output.

[0092] The aforementioned computer storage medium and terminal device are created based on the aforementioned adaptive welding method for workpieces. Their technical functions and beneficial effects will not be elaborated here. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A workpiece adaptive welding method, characterized in that, include: S1: Obtain the surface normal vector, weld direction tangent vector, and current weld position of each weld point on the workpiece surface within the scanning range of the acquisition device, so as to construct the absolute target pose matrix of the welding torch in the robot's global coordinate system and record the corresponding initial positioner external axis vector. S2: Obtain the global joint space vector of the current welding system, decouple to obtain the current external axis vector of the robotic arm, the joint vector of the robotic arm and the external axis vector of the positioner, calculate the external axis Jacobian submatrix and the robotic arm Jacobian submatrix respectively according to the current external axis position of the robotic arm and the angle of each joint of the robotic arm, and combine them to construct the global Jacobian matrix; S3: Obtain the welding path prediction sequence of several frames backward from the current time, transform the current weld position to the robot's global coordinate system, and correct the current weld position according to the initial external axis vector of the positioner and the current external axis vector of the positioner. Calculate the condition number of the robotic arm Jacobian submatrix under each prediction configuration, obtain the minimum collision distance between the robot's prediction envelope and the tooling fixture point cloud, and determine whether the condition number of the robotic arm Jacobian submatrix is ​​less than the condition number threshold and whether the minimum collision distance is less than the set safety threshold. S4: If none of them are less than, lock the current external axis vector of the robotic arm to solve the angular increment of each joint of the robotic arm, and correct the corresponding joint vector of the robotic arm according to the angular increment of each joint, weld the current weld position, and return to step S3. S5: If any is less than, unlock the current external axis vector of the robotic arm, calculate the Mohr-Ponrose pseudo-inverse of the global Jacobian matrix, and construct the null projection operator of the global joint space. With the constraint that the welding torch tip does not deviate from the absolute trajectory of the visual planning, the set redundancy elimination equation is executed to solve for the external axis velocity and joint velocity of the robotic arm. The external axis vector and joint vector of the robotic arm are then corrected, and welding is performed on the current weld position. Then, the process returns to step S3.

2. The method according to claim 1, characterized in that, The steps for constructing the absolute target pose matrix of the welding torch in the robot's global coordinate system include: Surface point clouds are collected, and voxel filtering and outlier removal are performed to obtain point clouds of the weld area; Surface fitting is performed on the point cloud of the weld area to obtain the surface normal vector at the weld point, and B-spline curve fitting is performed on the weld trajectory points to obtain the weld direction tangent vector. A local orthogonal rotation matrix is ​​constructed based on the surface normal vector and the weld direction tangent vector. Combined with the preset process push angle and working angle, the absolute target pose matrix between the robot end effector and each weld point is constructed.

3. The method according to claim 1, characterized in that, The steps for calculating the angular increments of each joint of the robotic arm and correcting the corresponding joint vectors based on these angular increments include: Set the speed of the robot arm's external axis to zero to lock the current external axis of the robot arm; Construct a selection mask matrix to eliminate the partial derivative space of the robot arm's external axis in the global Jacobian matrix, and obtain the robot arm Jacobian submatrix; The Jacobian submatrix of the robotic arm is projected onto the low-dimensional active axis subspace to obtain the dimension-reduced active axis Jacobian matrix. The angular increments of each joint of the robotic arm are calculated based on the Jacobian matrix of the reduced active axis, and the corresponding joint vectors of the robotic arm are corrected based on the angular increments of each joint.

4. The method according to claim 1, characterized in that, The steps to obtain the external axis velocity and joint velocity of the robotic arm include: The TCP velocity vector is determined based on the absolute target pose matrix. The product of the Morpenrose pseudoinverse and the TCP velocity vector is obtained to get the tracking term velocity. Obtain the product matrix of the Morpenrose pseudoinverse and the Jacobian matrix, and derive the null space projection operator based on the difference between the identity matrix and the product matrix. The product of the null space projection operator and the optimal gradient vector is obtained to determine the null space optimization speed. Summing the tracking term velocity and the zero-space optimization velocity yields the global joint velocity, and decoupling the external axis velocity and the joint velocity of the robotic arm is then obtained.

5. The method according to claim 4, characterized in that, The steps to obtain the optimal gradient vector include: Obtain the obstacle repulsion distance index, construct a first objective function with the goal of maximizing the obstacle repulsion distance, and solve for the optimal gradient value of the obstacle repulsion distance index; Obtain the kinematic operability index, construct an objective function that maximizes the kinematic operability, and solve for the optimal gradient value of the kinematic operability index. Obtain the joint limit avoidance index, construct an objective function that maximizes the joint limit avoidance performance, and solve for the optimal gradient value of the joint limit avoidance index. Based on the set weighting coefficients, the optimal gradient values ​​of the obstacle rejection distance index, kinematic maneuverability index, and joint limit avoidance index are weighted and summed to obtain the optimal gradient vector.

6. The method according to claim 5, characterized in that, The steps for obtaining the optimal gradient value of the obstacle repulsion distance index include: Calculate the minimum Euclidean distance between the robot's link surface and the tooling fixture point cloud in the current configuration, and determine whether the minimum Euclidean distance is less than or equal to the set safety threshold. If so, then construct the obstacle repulsion distance index function based on the set potential field gain coefficient, minimum Euclidean distance and set safety threshold, and take the derivative of the obstacle repulsion distance index function with respect to each joint variable to obtain the optimal gradient value of the obstacle repulsion distance index. If not, the optimal gradient value of the obstacle repulsion distance index is the zero vector.

7. The method according to claim 5, characterized in that, The steps to obtain the optimal gradient value of the kinematic operability index include: Calculate the determinant of the product matrix of the robot's global Jacobian matrix and its transpose, and take the square root to obtain the kinematic maneuverability index function. The optimal gradient value of the kinematic operability index is obtained by taking the partial derivatives of the kinematic operability index function with respect to each joint variable.

8. The method according to claim 5, characterized in that, The steps to obtain the optimal gradient value of the joint constraint avoidance index include: Obtain the upper physical limit angle, lower physical limit angle, and current angle of each joint, and calculate the midpoint coordinates of the stroke of each joint to construct a penalty function for joint deviation from the center position, and take the negative value to obtain the joint limit avoidance index function. The optimal gradient value of the joint limit avoidance index is obtained by taking the partial derivative of the joint limit avoidance index function with respect to each joint variable.

9. A computer storage medium, characterized in that, It stores executable program code; the executable program code is used to perform the workpiece adaptive welding method according to any one of claims 1 to 8.

10. A terminal device, characterized in that, It includes a memory and a processor; the memory stores program code that can be executed by the processor; the program code is used to execute the workpiece adaptive welding method according to any one of claims 1 to 8.

Images