Industrial robot welding track real-time optimization method and system based on deep learning

Abstract

The invention provides an industrial robot welding track real-time optimization method and system based on deep learning, and relates to the technical field of industrial robots, and the method comprises the steps: collecting three-dimensional point cloud data, carrying out the multi-scale processing to generate a high-precision curved surface, constructing a welding seam feature descriptor, and carrying out the real-time optimization of the welding track of an industrial robot; and the mapping relation between the technological parameters and the welding posture is optimized on the basis of a multi-target reward function, the mapping relation is converted into an initial movement track, the track is dynamically optimized through a segmented self-adaptive optimization model and a variable structure filtering algorithm, and self-adaptive adjustment of the welding process is achieved. The welding precision and quality of the complex curved surface are improved, and the adaptability and stability of robot welding are enhanced.

Description

Technical Field

[0001] The present invention relates to industrial robot technology, and particularly to a real-time optimization method and system for industrial robot welding trajectories based on deep learning. Background Art

[0002] With the improvement of the automation level of the manufacturing industry, industrial robots are increasingly widely used in the welding field. Industrial robot welding technology can significantly improve production efficiency, welding quality, and the safety of the working environment, especially having irreplaceable advantages when dealing with complex curved surface welding tasks. Traditional industrial robot welding trajectory planning mainly relies on manual teaching or offline programming methods to complete welding tasks through pre-set paths and process parameters. With the development of deep learning and computer vision technologies, adaptive welding trajectory planning methods based on sensors and intelligent algorithms have received wide attention, which can adjust welding parameters and trajectories in real time according to the actual welding environment and workpiece characteristics, improving welding quality and adaptability.

[0003] When dealing with complex curved surface welding in the prior art, it is difficult to accurately obtain high-precision surface features and weld information. Traditional point cloud data processing methods have poor noise reduction and feature extraction effects when facing complex geometric shapes, resulting in insufficient accuracy in subsequent weld recognition and trajectory planning, and unable to meet the requirements of high-quality welding.

[0004] Existing welding trajectory optimization methods lack a joint optimization mechanism for welding process parameters and robot postures. Most methods regard the optimization of welding process parameters and trajectory planning as two independent processes, ignoring the complex interaction relationship between process parameters and welding postures, and it is difficult to achieve the smoothness and efficiency of the trajectory while ensuring welding quality.

[0005] Existing trajectory optimization algorithms generally adopt a globally unified optimization strategy and lack the adaptive optimization ability for different curvature regions and welding difficulties. This method cannot flexibly adjust the optimization strategy according to the different characteristics of the welding trajectory. Especially when dealing with complex structures such as variable cross-sections and special-shaped joints, it is difficult to balance welding quality and motion efficiency, affecting the final welding effect. Summary of the Invention

[0006] Embodiments of the present invention provide a real-time optimization method and system for industrial robot welding trajectories based on deep learning, which can solve the problems in the prior art.

[0007] In the first aspect of the embodiments of the present invention, a real-time optimization method for industrial robot welding trajectories based on deep learning is provided, including: Collecting three-dimensional point cloud data of a complex curved surface, performing noise reduction and feature enhancement on the three-dimensional point cloud data by using a multi-scale decomposition algorithm, and generating high-precision surface data by combining a grid reconstruction method with curvature constraints; Build a dynamic feature space on the high-precision surface data, extract the set of weld track points in the dynamic feature space, and construct a weld feature descriptor that fuses geometric features and topological relationships; Based on the weld feature descriptor, take welding current, voltage, and wire feeding speed as the joint optimization objectives, and iteratively calculate the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establish the mapping relationship between the optimal process parameter combination and the welding posture sequence; Map the mapping relationship to the robot joint space, use an improved fast collision detection algorithm for obstacle avoidance verification, and generate an initial motion trajectory in combination with dynamic constraints; Establish a segmented adaptive optimization model for the initial motion trajectory, divide the optimization interval according to the comprehensive evaluation indexes of trajectory curvature, acceleration change rate, and process requirements, and use a variable structure filtering algorithm to smooth the trajectory within the optimization interval to achieve the dynamic optimization of the initial motion trajectory; Convert the optimized initial motion trajectory into a robot control instruction to realize the adaptive adjustment of the welding process.

[0008] Using a multi-scale decomposition algorithm to denoise and enhance the features of the three-dimensional point cloud data, and generating high-precision surface data by combining a mesh reconstruction method with curvature constraints includes: Perform multi-scale decomposition on the three-dimensional point cloud data based on wavelet transform, calculate the local curvature change rate at each decomposition scale, and use the local curvature change rate to perform multi-scale denoising processing on the three-dimensional point cloud data to obtain denoised point cloud data; Calculate the local neighborhood covariance matrix for the denoised point cloud data, extract the eigenvalues and eigenvectors of the local neighborhood covariance matrix, and construct a local geometric feature descriptor containing curvature information and normal vector information based on the eigenvalues and the eigenvectors; Calculate the feature distance between the local geometric feature descriptors using a Gaussian kernel function, and perform feature enhancement and density equalization processing according to the feature distance in combination with the local point cloud density distribution to generate feature-enhanced point cloud data; Construct a local curvature tensor field on the feature-enhanced point cloud data, estimate the local curvature change according to the local curvature tensor field, convert the local curvature change into a curvature constraint condition, and determine the sampling strategy in combination with the point cloud density to generate a shape quality index that satisfies the local curvature change; Iteratively optimize and adjust the mesh structure by combining the shape quality index, the curvature constraint condition, and the point cloud fitting error until convergence, and then output the high-precision surface data.

[0009] Construct a local curvature tensor field on the feature-enhanced point cloud data, estimate the local curvature change based on the local curvature tensor field, convert the local curvature change into a curvature constraint condition, and combine the point cloud density to determine a sampling strategy to generate a shape quality index that satisfies the local curvature change, including: The local curvature tensor field includes the curvature components of each point in the three coordinate axis directions. Based on the curvature components, an eigenvalue equation is constructed, and the principal curvature values are obtained by solving the equation where the determinant of the eigenvalue equation is zero; Calculate the Gaussian curvature by multiplying the principal curvature values, calculate the mean curvature by taking the arithmetic mean of the principal curvature values, calculate the curvature gradient vector based on the partial derivatives of the Gaussian curvature and the mean curvature in the three coordinate axis directions, and obtain the local curvature change rate by taking the square root of the sum of the squares of the components of the curvature gradient vector; Compare the local curvature change rate with a preset curvature threshold, determine the constraint coefficient according to the comparison result. When the local curvature change rate satisfies the preset curvature threshold, subtract the preset curvature threshold from the local curvature change rate and divide by the smoothing factor, and use the opposite number as the power of the exponential function to calculate the constraint coefficient; Construct a spherical search region centered on each data point in the point cloud data according to the constraint coefficient, divide the number of points in the spherical search region by the area of the spherical search region to obtain the point density, and calculate the ratio of the standard deviation to the mean value of the point density to obtain the shape quality index.

[0010] Based on the weld feature descriptor, take the welding current, voltage, and wire feeding speed as the joint optimization objectives, iteratively calculate the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establish the mapping relationship between the optimal process parameter combination and the welding posture sequence, including: Determine the process parameter change range based on the weld feature descriptor, establish the process constraint conditions among the welding current, welding voltage, and wire feeding speed, and construct the welding current, welding voltage, and wire feeding speed as the joint optimization objectives; Calculate the penetration control factor according to the welding current, welding voltage, and wire feeding speed, calculate the weld forming index based on the geometric parameters of the weld cross-section, and construct the penetration control factor and the weld forming index as the process quality evaluation indicators; Collect the welding torch posture angle and robot joint position data, calculate the deviation value of the welding torch posture angle relative to the target posture, calculate the motion smoothness of the robot joint position data, and construct the deviation value and the motion smoothness as the motion performance evaluation indicators; The process quality evaluation index and the motion performance evaluation index are combined to form a multi-objective reward function, and the joint optimization objective is iteratively optimized based on the gradient information of the multi-objective reward function. When the optimization gradient is less than the preset gradient threshold, the optimal process parameter combination is output; Optimize the welding torch attitude sequence according to the optimal process parameter combination, establish the mapping relationship from the weld seam feature descriptor to the optimal process parameter combination and the welding torch attitude sequence, and realize the collaborative planning of process parameters and welding postures.

[0011] Calculate the deviation value of the welding torch attitude angle relative to the target attitude, calculate the motion smoothness of the robot joint position, and construct the deviation value and the motion smoothness as the motion performance evaluation index, including: Calculate the angular deviation between the welding torch attitude angle and the target attitude angle, multiply the angular deviation by the corresponding direction weight coefficient respectively and sum them to obtain a weighted deviation index, and the weighted deviation index reflects the importance of different direction deviations; Collect the position signals and speed signals of each joint of the robot, take the second derivative of the position signals to obtain the speed difference signal, calculate the absolute value of the speed difference signal to obtain the position fluctuation index, and the position fluctuation index characterizes the acceleration change of the joint motion; Take the first derivative of the speed signal to obtain the speed difference increment signal, calculate the absolute value of the speed difference increment signal to obtain the speed change rate, and combine the position fluctuation index and the speed change rate to obtain the motion smoothness index; Based on the weighted deviation index and the motion smoothness index, a motion performance evaluation index is constructed by using a performance weight coefficient, and the performance weight coefficient is used to balance the influence of attitude accuracy and motion smoothness.

[0012] Map the mapping relationship to the robot joint space, use an improved fast collision detection algorithm to perform obstacle avoidance verification, and generate an initial motion trajectory in combination with dynamic constraints, including: Convert the mapping relationship in the workspace to the robot joint space through inverse kinematics to obtain the joint angle sequence; Use an improved fast collision detection algorithm to perform obstacle avoidance verification. The improved fast collision detection algorithm includes: constructing a hierarchical axis-aligned bounding box tree structure based on the joint angle sequence, and updating the spatial position parameters of the axis-aligned bounding box tree structure in real time as the joint angle sequence changes; calculating the minimum distance between adjacent bounding boxes in the axis-aligned bounding box tree structure, comparing the minimum distance with a preset safety threshold, and marking the collision risk area when the minimum distance is less than the preset safety threshold; Build a robot dynamics model including the joint inertia matrix, velocity coupling matrix, and gravity term. Based on the robot dynamics model, set the position limit, velocity limit, acceleration limit, and driving torque limit of joint motion. Take the position limit, velocity limit, acceleration limit, and driving torque limit as constraint conditions, and generate an initial motion trajectory that meets the collision risk area according to the constraint conditions.

[0013] Establish a piecewise adaptive optimization model for the initial motion trajectory. Divide the optimization interval according to the comprehensive evaluation index of trajectory curvature, acceleration change rate, and process requirements. Use a variable structure filtering algorithm to smooth the trajectory within the optimization interval. The dynamic optimization of the initial motion trajectory includes: Calculate the curvature, acceleration, and process parameters of the initial motion trajectory. Combine the curvature, acceleration, and process parameters to form a comprehensive evaluation index, and construct an adaptive threshold function based on the comprehensive evaluation index; Divide the initial motion trajectory into multiple optimization intervals according to the value of the adaptive threshold function. The adaptive threshold function decays exponentially with the change of the comprehensive evaluation index. Determine the time point when the comprehensive evaluation index is greater than the value of the adaptive threshold function as the interval boundary point; Establish a variable structure filter within the optimization interval based on the interval boundary point. The variable structure filter includes a state transition matrix and a state vector; calculate the state estimate value based on the observed value of the state vector, and take the state estimate value as the optimized trajectory parameter; Calculate the acceleration integral value and the maximum trajectory deviation of the optimized trajectory parameter. When the acceleration integral value and the maximum trajectory deviation meet the preset parameter threshold, output the smoothed trajectory. When they do not meet the preset parameter threshold, return to adjust the parameters of the variable structure filter until an optimized trajectory that meets the smoothing requirements is generated.

[0014] In the second aspect of the embodiments of the present invention, a real-time optimization system for the welding trajectory of an industrial robot based on deep learning is provided, including: The first unit is used to collect three-dimensional point cloud data of a complex surface, denoise and enhance the features of the three-dimensional point cloud data using a multi-scale decomposition algorithm, and generate high-precision surface data by combining a mesh reconstruction method with curvature constraints; The second unit is used to establish a dynamic feature space on the high-precision surface data, extract the weld seam trajectory point set in the dynamic feature space, and construct a weld seam feature descriptor that fuses geometric features and topological relationships; The third unit is configured to use the weld seam feature descriptor to take the welding current, voltage, and wire feeding speed as the combined optimization objectives, iteratively calculate the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establish a mapping relationship between the optimal process parameter combination and the welding posture sequence; The fourth unit is configured to map the mapping relationship to the robot joint space, perform obstacle avoidance verification using an improved fast collision detection algorithm, and generate an initial motion trajectory in combination with dynamic constraints; The fifth unit is configured to establish a segmented adaptive optimization model for the initial motion trajectory, divide the optimization interval according to the comprehensive evaluation indexes of trajectory curvature, acceleration change rate, and process requirements, and perform trajectory smoothing using a variable structure filtering algorithm within the optimization interval to achieve dynamic optimization of the initial motion trajectory; The sixth unit is configured to convert the optimized initial motion trajectory into a robot control instruction to achieve adaptive adjustment of the welding process.

[0015] In a third aspect of the embodiments of the present invention, an electronic device is provided, including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0016] In a fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0017] The beneficial effects of the present application are as follows: By collecting three-dimensional point cloud data of a complex surface and performing multi-scale decomposition processing and mesh reconstruction, high-precision surface data can be generated, effectively improving the accuracy and stability of weld seam trajectory recognition, and laying a solid foundation for subsequent welding trajectory planning.

[0018] Based on the weld seam feature descriptor, a mapping relationship between process parameters and welding postures is established, and the optimal parameter combination is iteratively calculated through a multi-objective reward function, which can achieve adaptive optimization of process parameters while ensuring welding quality, and improve welding efficiency and the quality of welding joints.

[0019] By using a segmented adaptive optimization model and a variable structure filtering algorithm to dynamically optimize the initial motion trajectory, and combining an improved fast collision detection algorithm for obstacle avoidance verification, a smooth and continuous welding trajectory that meets the process requirements can be generated, effectively reducing robot vibration and impact, extending the service life of the equipment, and improving the stability and reliability of the welding process. Description of the Drawings

[0020] Figure 1 It is a schematic flowchart of the real-time optimization method for the welding trajectory of an industrial robot based on deep learning according to an embodiment of the present invention; Figure 2 It is a flowchart for optimizing the point cloud curvature feature based on multi-scale decomposition according to an embodiment of the present invention; Figure 3 It is a schematic diagram showing the relationship between the point cloud quantity and the reconstruction accuracy according to an embodiment of the present invention; Figure 4 It is a comparative analysis diagram of the robot joint accelerations according to an embodiment of the present invention; Figure 5 It is a flowchart of the adaptive segmented filtering for trajectory optimization according to an embodiment of the present invention. Specific embodiments

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0023] Figure 1 It is a schematic flowchart of the real-time optimization method for the welding trajectory of an industrial robot based on deep learning according to an embodiment of the present invention, as Figure 1 shown, the method includes: Collect three-dimensional point cloud data of a complex surface, use a multi-scale decomposition algorithm to denoise and enhance the features of the three-dimensional point cloud data, and generate high-precision surface data by combining a grid reconstruction method with curvature constraints; Establish a dynamic feature space on the high-precision surface data, extract the weld seam trajectory point set in the dynamic feature space, and construct a weld seam feature descriptor that fuses geometric features and topological relationships; Based on the weld seam feature descriptor, take the welding current, voltage, and wire feeding speed as joint optimization objectives, iteratively calculate the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establish a mapping relationship between the optimal process parameter combination and the welding posture sequence; Map the mapping relationship to the robot joint space, use an improved fast collision detection algorithm for obstacle avoidance verification, and generate an initial motion trajectory in combination with dynamic constraints; Establish a piecewise adaptive optimization model for the initial motion trajectory, divide the optimization interval according to the comprehensive evaluation indexes of trajectory curvature, acceleration change rate and process requirements, and use a variable structure filtering algorithm to smooth the trajectory within the optimization interval to achieve the dynamic optimization of the initial motion trajectory; Convert the optimized initial motion trajectory into a robot control instruction to achieve the adaptive adjustment of the welding process.

[0024] In an optional implementation manner, use a multi-scale decomposition algorithm to denoise and enhance the features of the three-dimensional point cloud data, and generate high-precision surface data by combining a mesh reconstruction method with curvature constraints, including: Perform multi-scale decomposition on the three-dimensional point cloud data based on wavelet transform, calculate the local curvature change rate at each decomposition scale, and use the local curvature change rate to perform multi-scale denoising processing on the three-dimensional point cloud data to obtain denoised point cloud data; Calculate the local neighborhood covariance matrix for the denoised point cloud data, extract the eigenvalues and eigenvectors of the local neighborhood covariance matrix, and construct a local geometric feature descriptor containing curvature information and normal vector information based on the eigenvalues and the eigenvectors; Calculate the feature distance between the local geometric feature descriptors using a Gaussian kernel function, and perform feature enhancement and density equalization processing according to the feature distance combined with the local point cloud density distribution to generate feature-enhanced point cloud data; Construct a local curvature tensor field on the feature-enhanced point cloud data, estimate the local curvature change according to the local curvature tensor field, convert the local curvature change into a curvature constraint condition, and determine a sampling strategy in combination with the point cloud density to generate a shape quality index that satisfies the local curvature change; Iteratively optimize and adjust the mesh structure by using the shape quality index, the curvature constraint condition and the point cloud fitting error until convergence, and then output high-precision surface data.

[0025] Obtain the three-dimensional point cloud data to be processed, and this point cloud data may have problems such as measurement noise and uneven sampling. Perform multi-scale decomposition on the three-dimensional point cloud data based on wavelet transform. Specifically, use three-dimensional discrete wavelet transform to decompose the point cloud into different frequency components. For example, Haar wavelet or DB4 wavelet can be used as the basis function to decompose the point cloud coordinates for 3 to 5 layers to obtain low-frequency approximation components and high-frequency detail components.

[0026] At each decomposition scale, calculate the local curvature change rate, that is, within the neighborhood of point p (usually select 10 to 20 nearest neighbor points), estimate the principal curvatures k1 and k2 by fitting a local quadratic surface, and then calculate the curvature change rate as |Δk1| + |Δk2|, where Δk1 and Δk2 respectively represent the curvature gradients along the principal directions.

[0027] Perform multi-scale noise reduction processing on the point cloud using the local curvature change rate. Specifically: when the curvature change rate of a certain point exceeds the threshold (such as 2 times the average value), attenuate the contribution of this point in the high-frequency component by 50% - 80%; for the region with low and stable curvature change rate, retain its high-frequency detail information. Reconstruct each processed frequency component through inverse wavelet transform to obtain the noise-reduced point cloud data.

[0028] Calculate the local neighborhood covariance matrix for the noise-reduced point cloud data. For each point p, select the neighborhood points within the radius r (r is usually 2 - 3 times the average sampling distance of the point cloud) to construct a 3×3 covariance matrix. Extract the eigenvalues λ1≥λ2≥λ3 of this covariance matrix and the corresponding eigenvectors v1, v2, v3, where v3 approximately represents the normal vector at point p.

[0029] Construct a local geometric feature descriptor based on the eigenvalues and eigenvectors. This descriptor includes: surface change index λ3 / (λ1 + λ2 + λ3), principal curvature ratio (λ1 - λ2) / (λ1 + λ2), normal vector v3, and curvature change along the principal direction. These features together form an 8 - 12 dimensional feature vector used to describe the local geometric characteristics at point p.

[0030] Calculate the feature distance between local geometric feature descriptors using the Gaussian kernel function. For points p and q, the feature distance d(p,q)=exp(-||fp - fq||² / σ²), where fp and fq are the feature descriptors of the two points respectively, and σ is the kernel width parameter (usually taking a value of 0.6 - 0.8 times the average distance of the feature vectors).

[0031] Perform feature enhancement and density equalization processing according to the feature distance combined with the local point cloud density distribution: in the region with a small feature distance (such as a distance less than 0.2), generate new points through interpolation to enhance the model details; in the region with low point cloud density but significant geometric features (such as edges, corners), increase the sampling point density to 1.5 - 2 times the average density; appropriately reduce the point density to 0.7 - 0.8 times the average density in the flat region. This processing generates feature-enhanced point cloud data, which has a more balanced point distribution while maintaining the original geometric features.

[0032] Construct a local curvature tensor field on the feature-enhanced point cloud data. For each point p, fit a quadratic surface based on the points in its neighborhood, and calculate the principal curvatures k1, k2 and the corresponding principal directions. Through the tensor interpolation method, the discrete curvature information is extended to a continuous curvature tensor field. Estimate the local curvature change according to the local curvature tensor field, calculate the magnitude and direction of the curvature gradient, and convert the curvature change into curvature constraint conditions: in high curvature regions (such as when k1 or k2 is greater than 3 times the average curvature), the mesh side length should not exceed 0.5 times the average sampling distance; in regions with drastic curvature changes (the magnitude of the curvature gradient is greater than the threshold), the mesh surface should be subdivided along the direction of the maximum curvature change.

[0033] Determine the sampling strategy in combination with the point cloud density. On the premise of ensuring the mesh quality, the sampling in high-density regions can be appropriately reduced, and the sampling points are increased by interpolation in low-density regions. Thus, shape quality indicators that satisfy the local curvature change are generated, including: mesh area uniformity, the adaptability of the mesh side length to the local curvature, and the size of the triangle interior angles (ideally close to 60°).

[0034] Adjust the mesh structure by iteratively optimizing the shape quality indicators, curvature constraint conditions and point cloud fitting error. The specific optimization process uses the energy minimization method: set the energy function E = w1E_shape + w2E_curvature + w3E_fitting, where E_shape represents the shape quality indicator of the mesh structure, E_curvature represents the curvature constraint condition, E_fitting represents the point cloud fitting error, and w1, w2, w3 are weight coefficients (typical values are 0.4, 0.3, 0.3). In each iteration, adjust the mesh through operations such as edge flipping, vertex movement, and face splitting to reduce the value of the energy function.

[0035] When the change rate of the energy function is less than the preset threshold (such as 0.001) or reaches the maximum number of iterations (such as 50 times), the optimization process converges and outputs high-precision surface data. This surface data is represented in the form of a triangular mesh, while retaining the geometric features and curvature distribution information of the point cloud.

[0036] Experiments show that for the point cloud data of mechanical part containing 5% Gaussian noise (about 500,000 points), after being processed by this method, the root mean square error of the surface is reduced from the original 0.35mm to 0.08mm, the curvature fidelity is increased by 78%, and the geometric features in the edge and detail regions are effectively retained.

[0037] Figure 2 The flow chart of the point cloud curvature feature optimization based on multi-scale decomposition for the embodiment of the present invention: This flow chart details a complete technical solution for point cloud data processing and optimization, mainly including five key steps. First, wavelet transform is used to perform multi-scale decomposition on the three-dimensional point cloud data. By calculating the local curvature change rate at different decomposition scales, noise reduction processing is carried out on the point cloud data. Secondly, the local neighborhood covariance matrix is calculated for the noise-reduced point cloud data, and the eigenvalues and eigenvectors are extracted from it. Based on these features, a local geometric feature descriptor containing curvature information and normal vector information is constructed. In the third step, the Gaussian kernel function is used to calculate the feature distance between the feature descriptors, and feature enhancement and density equalization processing are carried out in combination with the local point cloud density distribution. In the fourth step, a local curvature tensor field is constructed based on the feature-enhanced point cloud data. By calculating the local curvature change, it is transformed into a curvature constraint condition, and the sampling strategy is determined in combination with the point cloud density. Finally, the shape quality index, curvature constraint condition, and point cloud fitting error are used to iteratively optimize and adjust the mesh structure until the convergence condition is reached, and then high-precision surface data is output. This multi-step optimization processing method can effectively improve the quality and accuracy of point cloud data.

[0038] In an alternative implementation, a local curvature tensor field is constructed on the feature-enhanced point cloud data, the local curvature change is estimated according to the local curvature tensor field, the local curvature change is transformed into a curvature constraint condition, and the sampling strategy is determined in combination with the point cloud density. The generated shape quality index that satisfies the local curvature change includes: The local curvature tensor field contains the curvature components of each point in the three coordinate axis directions. Based on the curvature components, an eigenvalue equation is constructed, and the principal curvature value is obtained by solving the equation where the determinant of the eigenvalue equation is zero. The product of the principal curvature values is calculated to obtain the Gaussian curvature, the arithmetic mean of the principal curvature values is calculated to obtain the mean curvature, the partial derivatives of the Gaussian curvature and the mean curvature in the three coordinate axis directions are calculated to obtain the curvature gradient vector, and the square root of the sum of the squares of the components of the curvature gradient vector is calculated to obtain the local curvature change rate. The local curvature change rate is compared with a preset curvature threshold, and the constraint coefficient is determined according to the comparison result. When the local curvature change rate satisfies the preset curvature threshold, the local curvature change rate minus the preset curvature threshold is divided by the smoothing factor, and the opposite number is used as the power of the exponential function to calculate the constraint coefficient. According to the constraint coefficient, a spherical search region is constructed with each data point in the point cloud data as the center. The number of points in the spherical search region is divided by the area of the spherical search region to obtain the point density, and the ratio of the standard deviation to the mean value of the point density is calculated to obtain the shape quality index.

[0039] Receive the point cloud data after feature enhancement. These feature-enhanced point cloud data contain the three-dimensional coordinate information of each point and the surface normal vector information of each point. The system constructs a local curvature tensor field on these point cloud data, and this tensor field describes the degree of curvature of the point cloud surface in different directions.

[0040] For each point in the point cloud, the system constructs a local curvature tensor by analyzing the distribution of its neighboring points. Specifically, with the current point as the center, all points within a spherical region with a radius of 0.05 meters are selected as neighboring points. For these neighboring points, the system calculates their relative position vectors with respect to the center point, and then constructs a covariance matrix.

[0041] By performing eigen-decomposition on the covariance matrix, three eigenvectors and corresponding eigenvalues are obtained. These three eigenvectors respectively represent the principal directions of the local surface, and the eigenvalues reflect the curvature components in these directions. For example, for points in a planar region, one of the eigenvalues is close to zero; while for points in a spherical region, the three eigenvalues are of similar magnitude.

[0042] Based on the curvature components calculated above, the system constructs an eigenvalue equation. This equation is expressed as an equation where a third-order determinant is equal to zero. By solving this equation, the system obtains three eigenvalues, and the two largest eigenvalues are defined as the principal curvature values. For example, for a certain point, by solving the eigenvalue equation, eigenvalues 0.1, 0.08, and 0.01 may be obtained, then 0.1 and 0.08 are the principal curvature values of this point.

[0043] After obtaining the principal curvature values, the system calculates the Gaussian curvature and the mean curvature. The Gaussian curvature is the product of the two principal curvature values, and the mean curvature is the arithmetic mean of the two principal curvature values. Taking the above example, the Gaussian curvature of this point is 0.1 multiplied by 0.08, which is equal to 0.008, and the mean curvature is (0.1 + 0.08) divided by 2, which is equal to 0.09.

[0044] To evaluate the change of curvature, the system calculates the partial derivatives of the Gaussian curvature and the mean curvature in the three coordinate axis directions to form a curvature gradient vector. Specifically, the system selects neighboring points at a distance of 0.01 meters from the current point in the x, y, and z directions, calculates the Gaussian curvature and the mean curvature of these points, and then divides the difference between these values and the curvature value of the current point by the distance to obtain the curvature gradients in each direction. For example, the Gaussian curvature gradient of a certain point in the x direction may be 0.002, in the y direction is 0.003, and in the z direction is 0.001.

[0045] Taking the square root of the sum of the squares of each component of the curvature gradient vector, the local curvature change rate is obtained. Continuing with the above example, the local curvature change rate of this point is the square root of (the square of 0.002 plus the square of 0.003 plus the square of 0.001), which is approximately equal to 0.0037.

[0046] Compare the calculated local curvature change rate with a preset curvature threshold, and determine the constraint coefficient according to the comparison result. The preset curvature threshold can be set to 0.003. When the local curvature change rate is greater than the preset curvature threshold, the system subtracts the preset curvature threshold from the local curvature change rate, then divides it by a smoothing factor (such as 0.001), takes the opposite number as the power of the exponential function, and calculates the constraint coefficient.

[0047] According to the calculated constraint coefficient, the system constructs a spherical search area centered on each data point in the point cloud data. The smaller the constraint coefficient, the larger the radius of the spherical search area. For example, for a constraint coefficient of 0.497, the system may set the radius of the spherical search area to 0.1 meter; while for a point with a constraint coefficient close to 1, the system may set the radius to 0.05 meter.

[0048] Count the number of points in the spherical search area, and divide the number of points by the surface area of the spherical search area to obtain the point density. For example, if there are 314 points in a spherical area with a radius of 0.1 meter, then the point density is 314 divided by (4π times the square of 0.1), which is approximately 250 points per square meter.

[0049] Calculate the ratio of the standard deviation to the average value of the point densities of all points to obtain the shape quality index. For example, if the average value of the point density is 200 points per square meter and the standard deviation is 40 points per square meter, then the shape quality index is 40 divided by 200, which is equal to 0.2. This index reflects the uniformity of the point cloud sampling and the adaptability to local curvature changes. A lower shape quality index indicates that the point cloud sampling is more uniform and better adapts to curvature changes.

[0050] Through the above steps, the system can adaptively adjust the sampling strategy according to the local curvature characteristics of the point cloud, increase the sampling density in areas with large curvature changes, and reduce the sampling density in areas with small curvature changes, so as to generate a shape quality index that meets the local curvature changes, providing a basis for subsequent point cloud processing and analysis.

[0051] Figure 3 Schematic diagram of the relationship between the number of point clouds and the reconstruction accuracy in the embodiments of the present invention: This figure shows the comparison of the accuracy performance of three different methods during the point cloud reconstruction process. The horizontal axis in the figure represents the number of point clouds (ranging from 102 to 107), and the vertical axis represents the reconstruction accuracy error (unit: mm). The three methods are represented by lines of different colors: the present technical solution (triangles), Poisson reconstruction (circles), and RANSAC fitting (squares). From the data trend, as the number of point clouds increases, the reconstruction errors of all three methods show a downward trend. The present technical solution performs the best, with the error decreasing from 0.032 mm at 102 points to 0.006 mm at 107 points; the Poisson reconstruction method has the largest error, decreasing from 0.068 mm at 102 points to 0.029 mm at 107 points; the RANSAC fitting method performs between the two, decreasing from 0.053 mm at 102 points to 0.019 mm at 107 points. Overall, the present technical solution shows better stability and lower error levels when the number of point clouds increases. Especially when the number of point clouds exceeds 105, the advantage of its reconstruction accuracy is more obvious, indicating that this solution has better adaptability and accuracy in processing large-scale point cloud data.

[0052] In an alternative embodiment, based on the weld seam feature descriptor, the welding current, voltage, and wire feeding speed are taken as the joint optimization objectives, and the optimal process parameter combination and welding posture sequence are iteratively calculated through a multi-objective reward function. Establishing the mapping relationship between the optimal process parameter combination and the welding posture sequence includes: Determine the process parameter change range based on the weld seam feature descriptor, establish the process constraint conditions among the welding current, welding voltage, and wire feeding speed, and construct the welding current, welding voltage, and wire feeding speed as the joint optimization objectives; Calculate the penetration control factor according to the welding current, welding voltage, and wire feeding speed, calculate the weld formation index based on the geometric parameters of the weld cross-section, and construct the penetration control factor and the weld formation index as the process quality evaluation indicators; Collect the welding torch posture angle and robot joint position data, calculate the deviation value of the welding torch posture angle relative to the target posture, calculate the motion smoothness of the robot joint position data, and construct the deviation value and the motion smoothness as the motion performance evaluation indicators; Form a multi-objective reward function with the process quality evaluation indicators and the motion performance evaluation indicators, iteratively optimize the joint optimization objectives based on the gradient information of the multi-objective reward function, and output the optimal process parameter combination when the optimization gradient is less than the preset gradient threshold; Optimize the welding torch posture sequence according to the optimal process parameter combination, establish the mapping relationship from the weld seam feature descriptor to the optimal process parameter combination and the welding torch posture sequence, and realize the collaborative planning of process parameters and welding postures.

[0053] Determine the process parameter variation range based on the weld feature descriptor, establish the process constraint conditions among the welding current, welding voltage, and wire feeding speed, and construct the welding current, welding voltage, and wire feeding speed as the joint optimization objectives.

[0054] For a V-shaped weld, its feature descriptors include the groove angle, root face size, and plate thickness information. When the plate thickness is 10 mm, the groove angle is 60°, and the root face size is 2 mm, according to the empirical data of Q235 steel welding materials, determine the variation range of the welding current to be 180 A to 240 A, the variation range of the welding voltage to be 22 V to 28 V, and the variation range of the wire feeding speed to be 6 m / min to 10 m / min. Establish the process constraint conditions: the welding current is proportional to the wire feeding speed, that is, for every 1 m / min increase in the wire feeding speed, the welding current increases by approximately 15 A; the welding voltage has a weak correlation with the welding current, and for every 1 V increase in the voltage, the welding current increases by approximately 5 A. These process constraint conditions form a three-dimensional parameter space as the search range for joint optimization.

[0055] Calculate the penetration control factor based on the welding current, welding voltage, and wire feeding speed, calculate the weld formation index based on the geometric parameters of the weld cross-section, and construct the penetration control factor and the weld formation index as the process quality evaluation indicators. The penetration control factor is calculated by dividing the product of the welding current and welding voltage by the wire feeding speed.

[0056] For example, when the welding current is 220 A, the welding voltage is 25 V, and the wire feeding speed is 8 m / min, the penetration control factor is 687.5. The larger this value, the deeper the penetration. Practice shows that for a V-shaped weld with a 10-mm plate thickness, the ideal value range of the penetration control factor is 650 to 700. The weld formation index is measured by the ratio of the weld reinforcement to the weld width, and the ideal weld formation index should be controlled between 0.1 and 0.2. When the welding parameters are within the above range and the actually measured weld reinforcement is 2 mm and the weld width is 15 mm, the weld formation index is 0.133, which is within the ideal range.

[0057] Collect the welding torch attitude angle and robot joint position data, calculate the deviation value of the welding torch attitude angle relative to the target attitude, calculate the motion smoothness of the robot joint position data, and construct the deviation value and the motion smoothness as the motion performance evaluation indicators.

[0058] The welding torch attitude angle is obtained by collecting the rotation matrix at the end of the welding torch in real time, including the working feed angle and the swing angle. For a V-shaped weld, the ideal working feed angle is 15° and the swing angle is 0°. When the actual working feed angle is 18° and the swing angle is 2°, the attitude deviation value is 5° (calculated by Euclidean distance). The robot joint position data includes the angle values of six joints. By calculating the change rate of the joint angular velocity at adjacent time points, the smoothness of the joint movement is obtained. Experimental data shows that when the change rate of the joint angular velocity is lower than 5° / s², the welding movement smoothness is good, and the corresponding evaluation value is 0.9 (full score is 1).

[0059] The process quality evaluation index and the motion performance evaluation index are combined to form a multi-objective reward function. Based on the gradient information of the multi-objective reward function, the joint optimization objective is iteratively optimized. When the optimization gradient is less than the preset gradient threshold, the optimal process parameter combination is output. The construction method of the multi-objective reward function is as follows: Multiply the evaluation value of the penetration control factor (full score is 1) by a weight of 0.4, multiply the evaluation value of the weld formation index (full score is 1) by a weight of 0.3, multiply the evaluation value of the attitude deviation value (full score is 1) by a weight of 0.2, and multiply the evaluation value of the motion smoothness by a weight of 0.1. The weighted sum of the four forms the final reward value. For example, R = 0.4P d + 0.3P f + 0.2P p + 0.1P s where R is the final comprehensive reward value, and the value range is [0,1]; P d is the evaluation value of the penetration control factor; P f is the evaluation value of the weld formation index; P p represents the evaluation value of the attitude deviation value; P s represents the evaluation value of the motion smoothness.

[0060] The calculation formula for the evaluation value of the penetration control factor is as follows: P d = exp(-|h a - h d | / h d ); where h a is the actual penetration value, which is calculated from the current (I), voltage (U) and wire feeding speed (v), and h d is the target penetration value; The calculation formula for the evaluation value of the weld formation index is as follows: P f = exp(-|w / h - α| / α); where w is the weld width, h is the weld height, and α is the ideal weld width-to-height ratio (usually 1.2 - 1.5); The calculation formula for the evaluation value of the attitude deviation value is as follows: P p = exp(-θ / θ m ax); where θ is the angle between the welding torch and the workpiece normal, and θ m ax is the maximum allowable deviation angle (usually 15°); The evaluation formula for the motion smoothness is as follows: P s = exp(-|a| / a m ax); where a is the acceleration of the welding torch, and a m ax is the maximum allowable acceleration value.

[0061] The optimization process uses the gradient descent method for iterative calculation, and the preset gradient threshold is 0.01. After about 50 iterations, the optimal process parameter combination is obtained: the welding current is 225 A, the welding voltage is 24.5 V, the wire feeding speed is 8.5 m / min. At this time, the penetration control factor is 650.7, the weld formation index is 0.15, the attitude deviation value is 2°, the motion smoothness is 0.95, and the total reward value is 0.92.

[0062] Optimize the welding torch attitude sequence according to the optimal process parameter combination, establish the mapping relationship between the weld feature descriptor and the optimal process parameter combination and the welding torch attitude sequence, and realize the collaborative planning of process parameters and welding postures. The method for optimizing the welding torch attitude sequence is as follows: Based on the fixed optimal process parameters, by adjusting the three parameters of the working feed angle, swing angle, and welding torch height of the welding torch, find the attitude combination that makes the welding molten pool the most stable. For the above V-shaped weld, the optimal welding torch attitude parameters are: the working feed angle is 15°, the swing angle is 0°, and the welding torch height is 12 mm.

[0063] Distribute the attitude parameters evenly along the weld path to form a welding attitude sequence. The finally established mapping relationship is: when the weld feature descriptor is {plate thickness = 10 mm, groove angle = 60°, root face size = 2 mm}, the corresponding optimal process parameter combination is {welding current = 225 A, welding voltage = 24.5 V, wire feeding speed = 8.5 m / min}, and the corresponding optimal welding attitude sequence is {working feed angle = 15°, swing angle = 0°, welding torch height = 12 mm}. This mapping relationship can be stored in the database for guiding the welding work of similar feature welds.

[0064] In an alternative embodiment, calculate the deviation value of the welding torch attitude angle relative to the target attitude, calculate the motion smoothness of the robot joint position, and construct the deviation value and the motion smoothness into a motion performance evaluation index, including: Calculate the angular deviation between the torch attitude angle and the target attitude angle, multiply the angular deviation by the corresponding direction weight coefficients respectively and sum them to obtain a weighted deviation index, which reflects the importance of deviations in different directions; Collect the position signals and speed signals of each joint of the robot, take the second derivative of the position signals to obtain a speed difference signal, calculate the absolute value of the speed difference signal to obtain a position fluctuation index, which characterizes the acceleration change of joint movement; Take the first derivative of the speed signal to obtain a speed difference increment signal, calculate the absolute value of the speed difference increment signal to obtain a speed change rate, and combine the position fluctuation index and the speed change rate to obtain a motion smoothness index; Based on the weighted deviation index and the motion smoothness index, construct a motion performance evaluation index using a performance weight coefficient, which is used to balance the influence of attitude accuracy and motion smoothness.

[0065] Real-time collect the actual attitude angle information of the torch through the end position sensor of the welding robot, including pitch angle, roll angle and yaw angle. At the same time, obtain the target attitude angle at the corresponding moment from the welding trajectory planning module. For each attitude angle component, calculate the angle difference between the actual value and the target value to obtain the angular deviations in three directions. For example, assume that at a certain moment, the actual pitch angle of the torch is 32.5 degrees and the target pitch angle is 30.0 degrees, then the pitch angle deviation is 2.5 degrees; the actual roll angle is 45.2 degrees and the target roll angle is 45.0 degrees, then the roll angle deviation is 0.2 degrees; the actual yaw angle is 88.7 degrees and the target yaw angle is 90.0 degrees, then the yaw angle deviation is 1.3 degrees.

[0066] Since the attitude deviations in different directions have different influences on the welding quality during welding, it is necessary to introduce direction weight coefficients. According to the weld type and process requirements, set the pitch angle weight coefficient to 0.5, the roll angle weight coefficient to 0.3, and the yaw angle weight coefficient to 0.2. Multiply the angular deviations in each direction by the corresponding weight coefficients and sum them to obtain a weighted deviation index. Taking the above data as an example, the weighted deviation index is calculated as: 2.5×0.5 + 0.2×0.3 + 1.3×0.2 = 1.51. This index comprehensively reflects the overall deviation degree of the torch attitude and at the same time reflects the importance difference of deviations in different directions.

[0067] The evaluation of the motion smoothness of the robot joints is divided into two parts: the position fluctuation index and the speed change rate. For the position fluctuation index, the system collects the position signals of the six joints of the robot at a frequency of 100 Hz, denoted as the position sequences of joint 1 to joint 6. Calculate the second derivative for each joint position sequence, that is, first calculate the position difference between adjacent two points to obtain the speed sequence, and then calculate the difference of the speed sequence to obtain the speed difference signal. After taking the absolute value of the speed difference signal, calculate the average value as the position fluctuation index of this joint. For example, the position fluctuation indexes of the six joints of the robot are 0.024, 0.015, 0.032, 0.008, 0.012, 0.005 (unit: degree / second²).

[0068] For the speed change rate index, the system directly collects the speed signals of each joint of the robot, and calculates the first derivative of the speed signal to obtain the speed difference increment signal. After taking the absolute value of the speed difference increment signal, calculate the average value as the speed change rate. For example, the speed change rates of the six joints of the robot are 0.118, 0.092, 0.145, 0.056, 0.074, 0.038 (unit: degree / second²).

[0069] Combine the position fluctuation index and the speed change rate to obtain the motion smoothness index. The specific method is to take the average value of the position fluctuation indexes of the six joints, getting 0.016; take the average value of the speed change rates of the six joints, getting 0.087. Then combine these two indexes according to the ratio of 7:3, and calculate the motion smoothness index as 0.016×0.7 + 0.087×0.3 = 0.0372. The smaller the index value, the smoother the robot's motion.

[0070] The final motion performance evaluation index is comprehensively constructed through the weighted deviation index and the motion smoothness index. Set the attitude accuracy weight coefficient to 0.6 and the motion smoothness weight coefficient to 0.4, then the motion performance evaluation index is calculated as: 1.51×0.6 + 0.0372×0.4 = 0.921. This comprehensive index realizes the comprehensive evaluation of the motion performance of the welding robot, and the setting of the weight coefficient reflects the balanced consideration of the attitude accuracy and the motion smoothness.

[0071] For different types of welding tasks, the weight coefficients of each item can be adjusted. For example, for precision welding tasks, the weight of the attitude accuracy in the final evaluation index can be increased to 0.8; for continuous long welds, the weight of the motion smoothness can be increased to 0.6. The system also provides a dynamic weight adjustment function, which automatically adjusts the weight coefficients of each item according to the real-time feedback during the welding process to further optimize the evaluation results.

[0072] In practical applications, the system realizes the optimization of the welding trajectory based on this evaluation index. When the evaluation index exceeds the preset threshold of 1.2, the trajectory optimization module is triggered, and the trajectory quality is improved by adjusting the path planning parameters or adding intermediate points. The optimized trajectory reduces the evaluation index from 1.35 to 0.87, significantly improving the welding quality and reducing the weld defect rate by 32%.

[0073] In addition, this evaluation method also supports two modes: offline analysis and online monitoring. Offline analysis is used for trajectory evaluation during the welding program development stage, and online monitoring is used for real-time monitoring during the production process, providing a basis for the dynamic adjustment of welding parameters and realizing the full-range quality guarantee of the welding process.

[0074] Figure 4 This is the comparative analysis chart of the robot joint accelerations in the embodiments of the present invention: This chart shows the comparison of the acceleration performances of three different schemes on six joints of the robot. The horizontal axis in the chart represents the joint numbers of the robot (from joint 1 to joint 6), and the vertical axis represents the acceleration change rate (rad / s 3 ). The three schemes are represented by different lines: the present technical scheme (triangle), the traditional trajectory planning (dot), and the quintic polynomial interpolation (square). From the data trend, the present technical scheme shows the lowest acceleration values at all joint positions, gradually decreasing from 0.032 rad / s at joint 1 3 to about 0.015 rad / s at joint 6 3 ; the acceleration values of the traditional trajectory planning scheme are the highest, decreasing from 0.085 rad / s at joint 1 3 to 0.060 rad / s at joint 6 3 ; the performance of the quintic polynomial interpolation scheme is between the two, decreasing from 0.052 rad / s at joint 1 3 to 0.035 rad / s at joint 6 3 . Overall, the acceleration values of the three schemes all show a gradually decreasing trend from joint 1 to joint 6, but the present technical scheme shows obvious advantages in reducing joint accelerations, which helps to improve the smoothness and accuracy of the robot movement.

[0075] In an alternative embodiment, mapping the mapping relationship to the robot joint space, and using an improved fast collision detection algorithm to perform obstacle avoidance verification, and generating an initial motion trajectory in combination with dynamic constraints includes: Converting the mapping relationship in the workspace to the robot joint space through inverse kinematics to obtain a joint angle sequence; Use an improved fast collision detection algorithm for obstacle avoidance verification. The improved fast collision detection algorithm includes: constructing a hierarchical axis-aligned bounding box tree structure based on the joint angle sequence, and updating the spatial position parameters of the axis-aligned bounding box tree structure in real time as the joint angle sequence changes; calculating the minimum distance between adjacent bounding boxes in the axis-aligned bounding box tree structure, comparing the minimum distance with a preset safety threshold, and marking the collision risk area when the minimum distance is less than the preset safety threshold. Establish a robot dynamics model including the joint inertia matrix, velocity coupling matrix, and gravity term. Based on the robot dynamics model, set the position limit, velocity limit, acceleration limit, and driving torque limit of joint movement. Use the position limit, velocity limit, acceleration limit, and driving torque limit as constraint conditions, and generate an initial motion trajectory that satisfies the collision risk area according to the constraint conditions.

[0076] In a robot motion planning method, it is necessary to convert the mapping relationship in the workspace to the joint space, perform collision detection and obstacle avoidance verification, and generate an initial motion trajectory considering dynamic constraints. Specifically, this method is implemented by the following technical means: For the mapping conversion between the workspace and the joint space, the system converts the path point sequence in the workspace into a joint angle sequence through inverse kinematics. In practical applications, assume that the end effector of a six-axis robotic arm needs to move from position point P1(500mm, 300mm, 200mm) to position point P2(600mm, 350mm, 250mm). The system first generates intermediate path points between these two points, such as P_mid(550mm, 325mm, 225mm), to form a path point sequence {P1, P_mid, P2}. Then, apply the inverse kinematics algorithm to each path point to obtain the corresponding joint angle sequence. For example, the joint angles corresponding to point P1 are {30°, 45°, -20°, 15°, 60°, 10°}, the joint angles corresponding to point P_mid are {35°, 48°, -18°, 18°, 62°, 12°}, and the joint angles corresponding to point P2 are {40°, 50°, -15°, 20°, 65°, 15°}, thus forming a complete joint angle sequence.

[0077] For collision detection and obstacle avoidance verification, this method uses an improved fast collision detection algorithm. This algorithm first constructs a hierarchical axis-aligned bounding box (AABB) tree structure based on the joint angle sequence. In actual implementation, the system represents each link part of the robot as a basic AABB, and then organizes these basic AABBs into a tree structure. For example, for a six-axis robotic arm, the base corresponds to AABB_0, the first link corresponds to AABB_1, and so on until the end effector corresponds to AABB_6. Each AABB is represented by its minimum vertex and maximum vertex coordinates. For example, the parameters of AABB_1 may be {minimum point (-100mm, -50mm, 0mm), maximum point (100mm, 50mm, 300mm)}.

[0078] As the joint angle sequence changes, the system updates the spatial position parameters of the AABB tree structure in real time. When the joint angles change from {30°, 45°, -20°, 15°, 60°, 10°} to {35°, 48°, -18°, 18°, 62°, 12°}, the system calculates the new position and orientation of each link according to forward kinematics and updates the spatial position parameters of the corresponding AABB accordingly. For example, AABB_2 may be updated to {minimum point (-80mm, -60mm, 280mm), maximum point (120mm, 40mm, 580mm)} under the new joint angles.

[0079] During collision detection, the system calculates the minimum distance between adjacent bounding boxes in the AABB tree structure. In specific implementation, for any two AABBs, the system first determines whether they overlap. If they do not overlap, it calculates their minimum distance. For example, the minimum distance between AABB_2 and the obstacle AABB_obs in the environment {minimum point (200mm, -100mm, 300mm), maximum point (300mm, 100mm, 500mm)} is calculated to be 80mm. The system compares this minimum distance with the preset safety threshold of 100mm. Since 80mm is less than 100mm, this position is marked as a collision risk area.

[0080] To generate an initial motion trajectory that satisfies dynamic constraints, the system first establishes a robot dynamic model that includes the joint inertia matrix, velocity coupling matrix, and gravity terms. In practical applications, the system can obtain the mass, center of mass position, and inertia tensor of each link through the CAD parameters of the robot. For example, the mass of the first link is 5kg, the center of mass position is (0mm, 0mm, 150mm), and the principal axis moment of inertia is (0.3kg·m², 0.3kg·m², 0.1kg·m²).

[0081] Based on the dynamic model, the system sets various limit constraints for joint movement. For the position limit, for example, the movement range of joint 1 is set to [-170°, 170°]; the speed limit is set to the maximum rotational speed of each joint, such as 120° / s for joint 1; the acceleration limit is set to the maximum acceleration of each joint, such as 180° / s² for joint 1; the driving torque limit is set to the maximum output torque of each motor, such as 150 N·m for joint 1.

[0082] After marking the collision risk areas, the system generates an initial motion trajectory according to the above constraints. For the positions marked as collision risk areas, the system will modify the original path points and add obstacle avoidance points.

[0083] For example, in the original path P1→P_mid→P2, if there is a collision risk at P_mid, the system will generate a new obstacle avoidance point P_avoid(550mm, 325mm, 300mm), making the new path become P1→P_avoid→P2.

[0084] Convert the new path into a joint angle sequence and verify whether it meets all dynamic constraints. If some points do not meet the constraints, the system will adjust the time parameters, reduce the speed and acceleration until all constraints are met, and finally generate a feasible initial motion trajectory.

[0085] Through the comprehensive application of the above technical means, this method can effectively convert the workspace mapping relationship to the joint space, achieve fast collision detection and obstacle avoidance, and generate a safe and feasible initial motion trajectory of the robot while meeting the dynamic constraints.

[0086] In an alternative embodiment, a segmented adaptive optimization model is established for the initial motion trajectory, and the optimization intervals are divided according to the comprehensive evaluation index of trajectory curvature, acceleration change rate and process requirements. The variable structure filtering algorithm is used for trajectory smoothing within the optimization intervals, and the dynamic optimization of the initial motion trajectory includes: Calculate the curvature, acceleration and process parameters of the initial motion trajectory, form a comprehensive evaluation index with the curvature, the acceleration and the process parameters, and construct an adaptive threshold function based on the comprehensive evaluation index; Divide the initial motion trajectory into multiple optimization intervals according to the value of the adaptive threshold function. The adaptive threshold function decays exponentially with the change of the comprehensive evaluation index, and determine the time points when the comprehensive evaluation index is greater than the value of the adaptive threshold function as the interval boundary points; A variable structure filter is established in the optimization interval based on the interval boundary point, wherein the variable structure filter includes a state transfer matrix and a state vector; a state estimation value is calculated based on the observed value of the state vector, and the state estimation value is used as the optimized trajectory parameter; The acceleration integral value and the maximum trajectory deviation of the optimized trajectory parameters are calculated, and a smooth trajectory is output when the acceleration integral value and the maximum trajectory deviation meet a preset parameter threshold. When the preset parameter threshold is not met, the parameters of the variable structure filter are adjusted until an optimized trajectory that meets the smoothness requirement is generated.

[0087] A segmented adaptive optimization model is established for the initial motion trajectory, and an optimization interval is divided according to comprehensive evaluation indicators of trajectory curvature, acceleration change rate and process requirements. A variable structure filtering algorithm is used to smooth the trajectory within the optimization interval.

[0088] Get the data of the initial motion trajectory, including the position coordinates, velocity, acceleration and other parameters at each time point. Calculate the curvature value of the initial motion trajectory, determine the radius of curvature through the circle formed by three adjacent points, and take its reciprocal as the curvature value. The acceleration change rate is obtained by calculating the derivative of the acceleration, and the process parameters are determined according to the application scenario, such as the feed rate and spindle speed in cutting processing. The curvature K, the acceleration change rate A and the process parameter P are combined to form a comprehensive evaluation index Q, which is calculated as follows: Q=w1×K+w2×A+w3×P, where w1, w2, and w3 are weight coefficients, which can be adjusted according to the actual application. For example, in a precision machining scenario, w1=0.4, w2=0.5, and w3=0.1 can be set.

[0089] Based on the comprehensive evaluation index, an adaptive threshold function T is constructed, T=T0×e^(-λ×Q), where T0 is the initial threshold and λ is the attenuation coefficient. Taking the motion trajectory of a precision machining center as an example, the initial threshold T0 is set to 5.0 and the attenuation coefficient λ is set to 0.2. When the comprehensive evaluation index Q in the area with drastic curvature changes reaches 3.5, the corresponding threshold function value T drops to 1.84, while when Q is 0.5 in the straight segment area, the threshold T remains at 4.09, realizing the adaptive change of the threshold with the motion state.

[0090] The initial trajectory is divided into multiple optimization intervals according to the comprehensive evaluation index. The comprehensive evaluation index Q of each time point is compared with the corresponding threshold function value T. When Q>T, the time point is marked as the interval boundary point. For example, the moments 0s, 2.5s, 4.8s, 7.2s, and 10s of the initial trajectory are identified as boundary points, thus dividing the 10-second trajectory into 4 optimization intervals.

[0091] A variable-structure filter is established within each optimization interval. The filter state transition matrix is constructed based on kinematic relationships. For a two-dimensional planar trajectory, the state vector includes position coordinates x, y, their first derivatives vx, vy, and second derivatives ax, ay. The filter parameters are adaptively adjusted according to the interval characteristics. In intervals with large curvature, a high position weight is set; in intervals with large velocity changes, a high velocity weight is set. Taking the interval from 0 to 2.5 s as an example, this interval has a relatively large curvature, with a position weight of 0.6, a velocity weight of 0.3, and an acceleration weight of 0.1. In the straight-line segment interval from 4.8 to 7.2 s, the position weight is set to 0.4, the velocity weight is 0.5, and the acceleration weight is 0.1.

[0092] Based on the sampling points of the initial trajectory as observations, the state estimation value is iteratively calculated through the variable-structure filter. For each time point t, the current state is predicted according to the state estimation value and the state transition matrix at the previous time point t - 1, and a more accurate state estimation is obtained by combining the observations. The state estimation value includes position, velocity, and acceleration, which are used as the optimized trajectory parameters. In the first optimization interval from 0 to 2.5 s, the maximum curvature of the original trajectory is 0.45, which is reduced to 0.32 after being processed by the variable-structure filter, while keeping the trajectory deviation within 0.15 mm.

[0093] Evaluate the quality of the optimized trajectory by calculating the integral value of acceleration S = ∫|a(t)|dt and the maximum trajectory deviation D = max|Poptimized - Poriginal|, and compare them with the preset parameter thresholds Smax and Dmax. If S ≤ Smax and D ≤ Dmax, the smoothed trajectory is output; otherwise, the filter parameters are adjusted and re-optimized. Taking machining as an example, the preset Smax = 25 m / s and Dmax = 0.2 mm. After the first optimization, S = 28.3 m / s and D = 0.18 mm are calculated. Although D meets the requirements, S exceeds the limit. Therefore, the filter parameters are adjusted, the position weight is reduced to 0.5, the velocity weight is increased to 0.4, and after recalculation, S = 23.6 m / s and D = 0.19 mm are obtained, meeting the preset requirements, and the optimized trajectory is output.

[0094] In practical applications, a certain numerical control machine tool performs a complex contour machining. The initial planned trajectory contains 300 points. After being processed by the above method, the acceleration change of the final trajectory is reduced by 47%, the maximum jerk is reduced from 12 m / s³ to 6.3 m / s³, and at the same time, the trajectory accuracy deviation is kept within 0.18 mm, meeting the machining accuracy requirements. Compared with traditional fixed-parameter filtering, the trajectory processed by the segmented adaptive optimization model maintains higher accuracy at the corners and better velocity smoothness in the straight-line segments, with a 35% improvement in comprehensive performance.

[0095] Figure 5 The flow chart of the adaptive segmented filtering for trajectory optimization in the embodiment of the present invention is as follows: The flow chart shows a complete trajectory optimization process. It starts from the initial motion trajectory, calculates the curvature, acceleration, and process parameters of the trajectory, constructs a comprehensive evaluation index based on these parameters, and establishes an adaptive threshold function. Subsequently, the system divides the trajectory into optimization intervals according to the adaptive threshold function and determines the boundary points of each interval. On this basis, a variable structure filter including a state transition matrix and a state vector is established, and this filter can adaptively adjust the state estimation. Next, the system calculates the state estimation value and uses it as the optimized trajectory parameter, and at the same time calculates the acceleration integral value and the maximum trajectory deviation. In the final stage of the process, a judgment mechanism is set: if the preset parameter threshold is met, the smooth trajectory is output as the final result; if the threshold requirement is not met, the filter parameters are returned for adjustment, and the state estimation calculation is performed again to form a closed-loop optimization process until an optimized trajectory that meets the requirements is obtained. This adaptive segmented optimization method can effectively handle the discontinuous points in the trajectory and ensure the smoothness and continuity of the optimization result.

[0096] In the second aspect of the embodiments of the present invention, a real-time optimization system for the welding trajectory of an industrial robot based on deep learning is provided, including: A first unit for collecting three-dimensional point cloud data of a complex surface, denoising and feature enhancing the three-dimensional point cloud data by using a multi-scale decomposition algorithm, and generating high-precision surface data by combining a grid reconstruction method with curvature constraints; A second unit for establishing a dynamic feature space on the high-precision surface data, extracting the weld seam trajectory point set in the dynamic feature space, and constructing a weld seam feature descriptor that fuses geometric features and topological relationships; A third unit for taking the welding current, voltage, and wire feeding speed as joint optimization objectives based on the weld seam feature descriptor, iteratively calculating the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establishing a mapping relationship between the optimal process parameter combination and the welding posture sequence; A fourth unit for mapping the mapping relationship to the robot joint space, performing obstacle avoidance verification by using an improved fast collision detection algorithm, and generating an initial motion trajectory in combination with dynamic constraints; A fifth unit for establishing a segmented adaptive optimization model for the initial motion trajectory, dividing the optimization interval according to the comprehensive evaluation index of the trajectory curvature, acceleration change rate, and process requirements, and performing trajectory smoothing within the optimization interval by using a variable structure filtering algorithm to achieve the dynamic optimization of the initial motion trajectory; A sixth unit for converting the optimized initial motion trajectory into a robot control instruction to achieve the adaptive adjustment of the welding process.

[0097] In the third aspect of the embodiments of the present invention, an electronic device is provided, including: A processor; A memory for storing processor-executable instructions; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0098] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0099] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for performing various aspects of the present invention are loaded.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time optimization method for the welding trajectory of an industrial robot based on deep learning, characterized in that Including: Collecting three-dimensional point cloud data of a complex surface, denoising and feature enhancing the three-dimensional point cloud data by using a multi-scale decomposition algorithm, and generating high-precision surface data by combining a mesh reconstruction method with curvature constraints; Establishing a dynamic feature space on the high-precision surface data, extracting a weld track point set in the dynamic feature space, and constructing a weld feature descriptor that fuses geometric features and topological relationships; Based on the weld feature descriptor, taking welding current, voltage, and wire feeding speed as joint optimization objectives, iteratively calculating the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establishing a mapping relationship between the optimal process parameter combination and the welding posture sequence; Mapping the mapping relationship to the robot joint space, performing obstacle avoidance verification by using an improved fast collision detection algorithm, and generating an initial motion trajectory in combination with dynamic constraints; Establishing a segmented adaptive optimization model for the initial motion trajectory, dividing the optimization interval according to a comprehensive evaluation index of trajectory curvature, acceleration change rate, and process requirements, and performing trajectory smoothing by using a variable structure filtering algorithm within the optimization interval to achieve dynamic optimization of the initial motion trajectory; Converting the optimized initial motion trajectory into a robot control instruction to achieve adaptive adjustment of the welding process.

2. The method according to claim 1, characterized in that Denoising and feature enhancing the three-dimensional point cloud data by using a multi-scale decomposition algorithm, and generating high-precision surface data by combining a mesh reconstruction method with curvature constraints includes: Performing multi-scale decomposition on the three-dimensional point cloud data based on wavelet transform, calculating the local curvature change rate at each decomposition scale, and performing multi-scale denoising processing on the three-dimensional point cloud data by using the local curvature change rate to obtain denoised point cloud data; Calculating the local neighborhood covariance matrix for the denoised point cloud data, extracting the eigenvalues and eigenvectors of the local neighborhood covariance matrix, and constructing a local geometric feature descriptor containing curvature information and normal vector information based on the eigenvalues and the eigenvectors; Calculating the feature distance between the local geometric feature descriptors by using a Gaussian kernel function, and performing feature enhancement and density equalization processing according to the feature distance in combination with the local point cloud density distribution to generate feature-enhanced point cloud data; Constructing a local curvature tensor field on the feature-enhanced point cloud data, estimating the local curvature change according to the local curvature tensor field, converting the local curvature change into a curvature constraint condition, and determining a sampling strategy in combination with the point cloud density to generate a shape quality index that satisfies the local curvature change; Iteratively optimizing and adjusting the mesh structure by using the shape quality index, the curvature constraint condition, and the point cloud fitting error until convergence, and then outputting high-precision surface data.

3. The method according to claim 2, wherein Constructing a local curvature tensor field on the feature-enhanced point cloud data, estimating the local curvature change according to the local curvature tensor field, converting the local curvature change into a curvature constraint condition, and determining a sampling strategy in combination with the point cloud density to generate a shape quality index that satisfies the local curvature change includes: The local curvature tensor field contains the curvature components of each point in the three coordinate axis directions. An eigenvalue equation is constructed based on the curvature components, and the principal curvature values are obtained by solving the equation where the determinant of the eigenvalue equation is zero; The product of the principal curvature values is calculated to obtain the Gaussian curvature, and the arithmetic mean of the principal curvature values is calculated to obtain the mean curvature. The curvature gradient vector is obtained by calculating the partial derivatives of the Gaussian curvature and the mean curvature in the three coordinate axis directions, and the local curvature change rate is obtained by taking the square root of the sum of the squares of the components of the curvature gradient vector; The local curvature change rate is compared with a preset curvature threshold, and the constraint coefficient is determined according to the comparison result. When the local curvature change rate meets the preset curvature threshold, the local curvature change rate minus the preset curvature threshold is divided by the smoothing factor, and the opposite number is used as the power of the exponential function to calculate the constraint coefficient; According to the constraint coefficient, a spherical search region is constructed with each data point in the point cloud data as the center. The number of points in the spherical search region is divided by the area of the spherical search region to obtain the point density, and the ratio of the standard deviation to the mean value of the point density is calculated to obtain the shape quality index.

4. The method according to claim 1, characterized in that, Based on the weld feature descriptor, the welding current, voltage, and wire feeding speed are used as the joint optimization objectives. The optimal process parameter combination and welding posture sequence are iteratively calculated through a multi-objective reward function, and the mapping relationship between the optimal process parameter combination and the welding posture sequence is established, including: Based on the weld feature descriptor, the variation range of the process parameters is determined, the process constraint conditions between the welding current, welding voltage, and wire feeding speed are established, and the welding current, welding voltage, and wire feeding speed are constructed as the joint optimization objectives; The penetration control factor is calculated according to the welding current, welding voltage, and wire feeding speed, and the weld forming index is calculated based on the geometric parameters of the weld cross-section. The penetration control factor and the weld forming index are constructed as the process quality evaluation indexes; The welding torch posture angle and robot joint position data are collected, the deviation value of the welding torch posture angle relative to the target posture is calculated, and the motion smoothness of the robot joint position data is calculated. The deviation value and the motion smoothness are constructed as the motion performance evaluation indexes; The process quality evaluation index and the motion performance evaluation index are combined to form a multi-objective reward function, and the joint optimization objective is iteratively optimized based on the gradient information of the multi-objective reward function. When the optimization gradient is less than the preset gradient threshold, the optimal process parameter combination is output; The welding torch posture sequence is optimized according to the optimal process parameter combination, and the mapping relationship from the weld feature descriptor to the optimal process parameter combination and the welding torch posture sequence is established to realize the collaborative planning of the process parameters and the welding posture.

5. The method according to claim 4, wherein Calculating the deviation value of the welding torch posture angle relative to the target posture, calculating the motion smoothness of the robot joint position, and constructing the deviation value and the motion smoothness as the motion performance evaluation indexes, including: Calculate the angular deviation between the welding torch attitude angle and the target attitude angle, multiply the angular deviation by the corresponding direction weight coefficients respectively and sum them to obtain a weighted deviation index, which reflects the importance of deviations in different directions; Collect the position signals and velocity signals of each joint of the robot, take the second derivative of the position signals to obtain a velocity difference signal, calculate the absolute value of the velocity difference signal to obtain a position fluctuation index, which characterizes the acceleration change of joint movement; Take the first derivative of the velocity signal to obtain a velocity difference increment signal, calculate the absolute value of the velocity difference increment signal to obtain a velocity change rate, and combine the position fluctuation index and the velocity change rate to obtain a motion smoothness index; Based on the weighted deviation index and the motion smoothness index, construct a motion performance evaluation index using a performance weight coefficient, and the performance weight coefficient is used to balance the influence of attitude accuracy and motion smoothness.

6. The method according to claim 1, wherein Map the mapping relationship to the robot joint space, and use an improved fast collision detection algorithm for obstacle avoidance verification, and generate an initial motion trajectory in combination with dynamic constraints, including: Convert the mapping relationship in the workspace to the robot joint space through inverse kinematics to obtain a joint angle sequence; Use an improved fast collision detection algorithm for obstacle avoidance verification. The improved fast collision detection algorithm includes: constructing a hierarchical axis-aligned bounding box tree structure based on the joint angle sequence, and updating the spatial position parameters of the axis-aligned bounding box tree structure in real time as the joint angle sequence changes; calculating the minimum distance between adjacent bounding boxes in the axis-aligned bounding box tree structure, and comparing the minimum distance with a preset safety threshold. When the minimum distance is less than the preset safety threshold, mark the collision risk area; Establish a robot dynamics model including a joint inertia matrix, a velocity coupling matrix, and a gravity term. Based on the robot dynamics model, set the position limit, velocity limit, acceleration limit, and driving torque limit of joint movement. Take the position limit, the velocity limit, the acceleration limit, and the driving torque limit as constraint conditions, and generate an initial motion trajectory that satisfies the collision risk area according to the constraint conditions.

7. The method according to claim 1, characterized in that, Establish a piecewise adaptive optimization model for the initial motion trajectory, divide the optimization interval according to a comprehensive evaluation index of trajectory curvature, acceleration change rate, and process requirements, and use a variable structure filtering algorithm to smooth the trajectory within the optimization interval to achieve dynamic optimization of the initial motion trajectory, including: Calculate the curvature, acceleration, and process parameters of the initial motion trajectory, combine the curvature, the acceleration, and the process parameters into a comprehensive evaluation index, and construct an adaptive threshold function based on the comprehensive evaluation index; Divide the initial motion trajectory into multiple optimization intervals according to the value of the adaptive threshold function. The adaptive threshold function decays exponentially as the comprehensive evaluation index changes. Determine the time point when the comprehensive evaluation index is greater than the value of the adaptive threshold function as the interval boundary point; A variable structure filter is established based on the interval boundary points within the optimization interval, and the variable structure filter includes a state transition matrix and a state vector; a state estimate value is calculated based on the observed value of the state vector, and the state estimate value is used as the optimized trajectory parameter. The acceleration integral value and the maximum trajectory deviation of the optimized trajectory parameter are calculated. When the acceleration integral value and the maximum trajectory deviation meet the preset parameter thresholds, a smooth trajectory is output. When they do not meet the preset parameter thresholds, the parameters of the variable structure filter are adjusted until an optimized trajectory that meets the smoothness requirements is generated.

8. An industrial robot welding trajectory real-time optimization system based on deep learning, which is used to implement the method described in any one of the foregoing claims 1-7, and is characterized in that It includes: The first unit is used to collect the three-dimensional point cloud data of the complex surface, perform noise reduction and feature enhancement on the three-dimensional point cloud data by using a multi-scale decomposition algorithm, and generate high-precision surface data by combining a mesh reconstruction method with curvature constraints. The second unit is used to establish a dynamic feature space on the high-precision surface data, extract the weld trajectory point set in the dynamic feature space, and construct a weld feature descriptor that fuses geometric features and topological relationships. The third unit is used to take the welding current, voltage, and wire feeding speed as the joint optimization objectives based on the weld feature descriptor, iteratively calculate the optimal process parameter combination and welding posture sequence through a multi-objective reward function, and establish a mapping relationship between the optimal process parameter combination and the welding posture sequence. The fourth unit is used to map the mapping relationship to the robot joint space, perform obstacle avoidance verification by using an improved fast collision detection algorithm, and generate an initial motion trajectory in combination with dynamic constraints. The fifth unit is used to establish a segmented adaptive optimization model for the initial motion trajectory, divide the optimization interval according to the comprehensive evaluation indexes of trajectory curvature, acceleration change rate, and process requirements, and perform trajectory smoothing by using a variable structure filtering algorithm within the optimization interval to realize the dynamic optimization of the initial motion trajectory. The sixth unit is used to convert the optimized initial motion trajectory into a robot control instruction to realize the adaptive adjustment of the welding process.

9. An electronic device, characterized in that, It includes: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is realized.

Images