Ultra-thin component multi-region reinforcement adaptive path planning method and device

Abstract

A method and apparatus for adaptive path planning for multi-region strengthening of ultrathin components in collaboration with a robotic arm are disclosed. The method involves using a binocular line-scanning laser 3D camera to acquire 3D data of the ultrathin component surface; employing a region partitioning algorithm to divide the 3D point cloud of the ultrathin component surface into regions; calculating the center point coordinates and estimating the normal vectors of the 3D point clouds of each region; using a genetic algorithm to plan the optimal access sequence for each region and determine the optimal processing order; completing the camera and robotic arm hand-eye calibration, as well as the calibration of the laser center point and laser direction; converting the center point coordinates and normal vectors of each region into the corresponding pose coordinates of the robotic arm; controlling the robotic arm to place each region of the ultrathin component sequentially into the processing area of ​​a 3D galvanometer according to the planned optimal access sequence and the pose coordinates, and then using the galvanometer to perform ultrafast laser shock strengthening on the component surface.

Description

Technical Field

[0001] This invention relates to the field of adaptive machining technology for industrial robots, and in particular to a method and apparatus for multi-regional reinforcement adaptive path planning of ultra-thin components. Background Technology

[0002] In modern industrial manufacturing, ultra-thin components such as aero-engine blades and blade sheaths operate under harsh environments with high temperatures, high pressures, and high speeds for extended periods. Therefore, their surface fatigue life and corrosion resistance are crucial to overall safety. Traditional material strengthening methods, such as mechanical shot peening or chemical treatment, have several limitations: firstly, these methods struggle to precisely control complex curved surfaces, easily leading to over- or under-treatment; secondly, the processes are complex, requiring multiple clamping and adjustments, significantly extending the production cycle.

[0003] Ultrafast laser shock blasting is an advanced non-contact surface modification technology. It uses nanosecond or femtosecond laser pulses to deliver high-energy-density shocks to the material surface, inducing microscopic plastic deformation and generating a favorable residual compressive stress field. This significantly improves the fatigue strength, corrosion resistance, and service life of components. Compared to traditional methods, ultrafast laser shock blasting offers advantages such as controllable depth of penetration, a small heat-affected zone, and easily adjustable process parameters. However, the widespread adoption of this technology faces the following key technical bottlenecks:

[0004] First, determining the automated machining path for complex curved workpieces is difficult. Traditional methods rely on offline programming, requiring operators to manually specify the machining trajectory. This is extremely labor-intensive and error-prone for ultra-thin components with multiple machining zones. An unreasonable access sequence between machining areas can lead to frequent, large-amplitude movements of the robotic arm, resulting in low machining efficiency. Second, the coordinate transformation accuracy of the vision-guided system directly affects machining quality. In an eye-to-hand configuration, the spatial relationship between the camera and the robotic arm's end effector needs precise calibration. Any calibration error will cause the actual machining position to deviate from the target, which is unacceptable, especially in scenarios where ultra-thin components have extremely high precision requirements. Third, the coordinated control of the 3D galvanometer system and the robotic arm requires a high degree of spatiotemporal synchronization. The galvanometer is used to finely adjust the laser's position within the micrometer range, while the robotic arm is responsible for coarse positioning; both must work closely together to achieve efficient machining. Current technology lacks a systematic solution to integrate these two aspects. Finally, a complete methodology from point cloud acquisition, geometric feature calculation, optimal path planning to coordinate transformation is still lacking.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To address the shortcomings or defects of the existing technology, a method and apparatus for adaptive path planning for multi-region strengthening of ultra-thin components is provided. By introducing visual guidance, point cloud data of the workpiece is acquired, processed, and intelligent path planning for laser strengthening is performed. This controls the laser to perform accurate laser strengthening, achieving automated and efficient processing of multi-region strengthening of ultra-thin components. This avoids the complexity of traditional offline programming and improves processing efficiency and workpiece quality.

[0007] The objective of this invention is achieved through the following technical solutions.

[0008] An adaptive path planning method for multi-region reinforcement of ultrathin components includes,

[0009] A binocular line-scanning laser 3D camera was used to acquire three-dimensional data of the surface of the ultrathin component, and a region division algorithm was used to divide the three-dimensional point cloud of the ultrathin component surface into regions.

[0010] The center point coordinates and normal vectors of the 3D point clouds of each region after regional division are calculated and estimated.

[0011] A genetic algorithm is used to plan the optimal access sequence for each partition and determine the optimal processing order.

[0012] Complete the calibration of the camera and robotic arm's hand-eye alignment, as well as the calibration of the laser center point and laser direction of origin;

[0013] Convert the center point coordinates and normal vectors of each partition into the corresponding pose coordinates of the robotic arm;

[0014] The robotic arm is controlled to place each section of the ultrathin component sequentially into the processing area of ​​the three-dimensional galvanometer according to the planned optimal access sequence and the pose coordinates. The galvanometer then performs ultrafast laser shock blasting on the surface of the component.

[0015] The method described includes calculating the center point coordinates and estimating the normal vectors of the 3D point clouds of each region after regional division.

[0016] Extract the 3D point cloud data of each partition; calculate the geometric center of the point cloud of each partition as the coordinates of the center point of that partition; calculate the normal vector of the surface of that partition based on the covariance matrix principal component analysis of the nearest neighbor points, and perform normal vector normalization.

[0017] In the method described, the calculation of the center point coordinates of each partition's 3D point cloud includes,

[0018] The point cloud of partition i is The center point is then calculated as follows: Calculate the distance from each point to the geometric center; points closer to the geometric center have a higher weight. Based on this, recalculate the weighted center. ,in , Let k be the Euclidean distance from the k-th point to the initial geometric center. For scale parameters;

[0019] For each point in the partitioned point cloud, the normal vector is calculated based on its nearest neighbors using principal component analysis: for each point... The set of k nearest neighbors is Calculate the covariance matrix of these points: ,in The mean of the nearest neighbors. For nearest neighbors, with respect to the covariance matrix: Eigenvalue decomposition is performed, and the eigenvector corresponding to the smallest eigenvalue is the surface normal vector at that point. The average or weighted average of the normal vectors of all points within the partition is then calculated to obtain the representative normal vector of that partition. Where N is the number of points in the partition. Let be the normal vector of the j-th point. Finally, normalize the normal vector so that its length is 1.

[0020] The method described above includes using a genetic algorithm to plan the optimal access sequence for each partition.

[0021] Using the center point of each partition as a path node, calculate multiple cost functions such as distance between any two partitions, angle between normal vectors, acceleration change, and workload; weight and fuse the cost functions to form a fitness evaluation function; use a genetic algorithm to iteratively search the population for the optimal partition access sequence that minimizes the total movement distance, smooths the turning, and balances the workload through selection, crossover, and mutation operations.

[0022] In the method described, the cost function is:

[0023] in The weight coefficients are the partition set. Each partition has a center point. and normal vector Any two partitions and The transfer cost between them is a weighted sum of multiple sub-costs, representing the distance the robotic arm moves: Cost of the angle between normal vectors: Acceleration cost: Workload cost: ;in Let j be the number of points in partition j. To avoid zero denominators for small positive numbers, the cost of maintaining regional continuity is: .

[0024] The method described includes camera and robotic arm hand-eye calibration,

[0025] The camera is fixed to the end effector fixture of the robotic arm in an eye-out manner, and calibration plate images are acquired in multiple poses during the movement of the robotic arm; the intrinsic parameters of the camera are obtained by Zhang Zhengyou calibration method; the transformation matrix between the camera coordinate system and the robotic arm wrist joint coordinate system is solved by Tsai-Lenz algorithm; laser calibration includes determining the position of the laser source center point in the robotic arm coordinate system and the direction vector of the laser beam.

[0026] In the method described, the conversion of the center point coordinates and normal vectors of each partition into the corresponding pose coordinates of the robotic arm includes,

[0027] The coordinates of the center point of the partition under the camera's field of view are transformed to the robot arm's base coordinate system using a hand-eye calibration matrix; the normal vector of the partition is transformed in the corresponding direction using a rotation matrix; the pose coordinates of the robot arm's end effector are calculated based on the transformed center point position and normal vector direction, and the pose coordinates include three-dimensional position coordinates and three Euler angles.

[0028] In the method described, ultrafast laser shock peening includes: impacting the surface of an ultrathin component region by region with nanosecond or femtosecond-level ultrafast laser pulses; and inducing microscopic plastic deformation on the material surface by the laser pulses.

[0029] An apparatus for implementing the aforementioned robotic arm-cooperative ultrathin component multi-region reinforcement adaptive path planning method includes,

[0030] A robotic arm system with at least six degrees of freedom to grip a camera and a laser output head;

[0031] A three-dimensional galvanometer scanning system that controls the scanning path of the laser on the surface of a workpiece;

[0032] A camera that provides real-time visual feedback to obtain high-precision 3D point cloud data of the workpiece surface;

[0033] The control system controls the robotic arm system, the three-dimensional galvanometer scanning system, the camera, and the laser output head to perform ultrafast laser shock strengthening based on the aforementioned robotic arm-coordinated ultrathin component multi-region strengthening adaptive path planning method.

[0034] In the device described above, the three-dimensional galvanometer scanning system includes a three-dimensional tunable acousto-optic deflector or dynamic reflector array for three-dimensional scanning of the laser spot position, and an imaging system for focusing the laser onto the workpiece surface.

[0035] Compared with existing technologies, the beneficial effects of this invention are as follows: Based on region division, this invention performs fine processing on the 3D point cloud of each partition, calculating the geometric center and surface normal vector of each partition. Unlike simply taking the arithmetic mean of the partition point cloud, this method uses a weighted distance search method to ensure the representativeness of the center point; simultaneously, principal component analysis is performed through the covariance matrix of neighboring points to obtain accurate surface normal vectors, laying the foundation for subsequent robotic arm pose calculation. This method breaks through traditional greedy or heuristic path planning methods. A genetic algorithm is used to transform the partition access order planning problem into a multi-objective optimization problem, comprehensively considering the Euclidean distance between partitions, the angle change of the normal vectors, the acceleration fluctuation of the robotic arm motion, the workload of each partition, and the spatial continuity of the partitions. Through evolutionary operations such as selection, crossover, and mutation, the algorithm can quickly converge to a near-global optimal path scheme, significantly outperforming traditional heuristic methods. Precise coordinate transformation for eye-to-hand vision guidance. This method performs detailed hand-eye calibration and laser calibration under an eye-to-hand configuration, establishing a precise transformation relationship from the camera's field of view coordinate system to the robotic arm's base coordinate system. Based on this, matrix operations reliably transform the center point and normal vector of the partition from the visual coordinate system to the robotic arm coordinate system, laying the foundation for the precise positioning of the robotic arm. This achieves the transformation from offline programming to online adaptive machining, improving work efficiency; through path optimization, it reduces the ineffective movement of the robotic arm, lowering energy consumption and processing time; it establishes a complete vision guidance and coordinate transformation system, improving machining accuracy and reliability; and it is applicable to ultra-thin components of different shapes and sizes.

[0036] The description provided is merely an overview of the technical solution of this invention. In order to make the technical means of this invention clearer and more understandable, so that those skilled in the art can implement it according to the contents of the specification, and to make the described and other objects, features and advantages of this invention more obvious and understandable, specific embodiments of this invention are described below. Attached Figure Description

[0037] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0038] In the attached diagram:

[0039] Figure 1This is a flowchart illustrating the method steps of a galvanometer-robotic arm collaborative method and apparatus for multi-regional reinforcement adaptive path planning of ultrathin components according to the present invention.

[0040] Figure 2 This is a structural schematic diagram of a galvanometer-robotic arm collaborative ultrathin component multi-region reinforcement adaptive path planning method and device according to the present invention.

[0041] Figure 3 This is a flowchart illustrating the calculation of the robotic arm reinforcement trajectory in a galvanometer-robotic arm collaborative adaptive path planning method and device for multi-region reinforcement of ultrathin components according to the present invention.

[0042] In the attached diagram,

[0043] 1-Robotic arm, 2-Workpiece fixture, 3-Ultra-thin component, 4-3D galvanometer scanning system, 5-Binocular line-scanning laser 3D camera, 6-Camera bracket, 7-Optical platform, 8-Control cabinet, 9-Laser, 10-Laser optical path.

[0044] The workpiece fixture 2 is equipped with an ultra-thin component 3, which is fixed to the end flange of the robotic arm 1. The optical platform 7 is equipped with a three-dimensional galvanometer scanning system 4, a binocular line-scanning laser 3D camera 5, a laser 9, and a laser optical path 10. The control cabinet contains the control box for the laser 9. The binocular line-scanning laser 3D camera is used to photograph the ultra-thin component 3 and acquire its point cloud data. The robotic arm 1 moves the ultra-thin component 3. The laser 9 generates a laser beam and outputs it through the laser optical path 10. The laser beam irradiates the ultra-thin component 3, and the three-dimensional galvanometer scanning system 4 dynamically scans the laser position. The robotic arm 1, with the ultra-thin component 3, continuously changes position to achieve laser strengthening of the entire outer surface of the ultra-thin component 3.

[0045] The present invention will be further explained below with reference to the accompanying drawings and embodiments. Detailed Implementation

[0046] Specific embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While specific embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0047] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0048] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments, and the accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0049] To better understand, such as Figures 1 to 3 As shown, the method includes,

[0050] A binocular line-scanning laser 3D camera was used to acquire three-dimensional data of the surface of the ultrathin component, and a region division algorithm was used to divide the three-dimensional point cloud of the ultrathin component surface into regions.

[0051] The center point coordinates and normal vectors of the 3D point clouds of each region after regional division are calculated and estimated.

[0052] A genetic algorithm is used to plan the optimal access sequence for each partition and determine the optimal processing order.

[0053] Complete the calibration of the camera and robotic arm's hand-eye alignment, as well as the calibration of the laser center point and laser direction of origin;

[0054] Convert the center point coordinates and normal vectors of each partition into the corresponding pose coordinates of the robotic arm;

[0055] The robotic arm is controlled to place each section of the ultrathin component sequentially into the processing area of ​​the three-dimensional galvanometer according to the planned optimal access sequence and the pose coordinates. The galvanometer then performs ultrafast laser shock blasting on the surface of the component.

[0056] In a preferred embodiment of the method, calculating the center point coordinates and estimating the normal vectors of the 3D point clouds of each region after region division includes:

[0057] Extract the 3D point cloud data of each partition; calculate the geometric center of the point cloud of each partition as the coordinates of the center point of that partition; calculate the normal vector of the surface of that partition based on the covariance matrix principal component analysis of the nearest neighbor points, and perform normal vector normalization.

[0058] In a preferred embodiment of the method, the three-dimensional point cloud region division of the surface of the ultrasonic component includes:

[0059] The region partitioning algorithm is designed for the region partitioning of the 3D point cloud on the surface of ultra-thin components. With "adapting to the low stiffness and high precision characteristics of the components" as its core, it executes the algorithm step-by-step according to the logic of "data definition - preprocessing - feature extraction - core segmentation - post-processing - result output". The specific process is as follows: First, the basic definition is clarified, assuming the 3D point cloud on the surface of the ultra-thin component is... [ Let i be the three-dimensional coordinates of the i-th point, and let some points contain normal vectors. The core objective of the algorithm is to divide P into non-overlapping and completely covered sub-point clouds. (satisfy (where U represents the union), while preserving the surface deformation and microscopic features of the component; in the preprocessing stage, firstly, statistical filtering is used to remove noise, taking K=15 (where K is: the "nearest neighbor number parameter" in the K-nearest neighbor algorithm; 15 means: the 15 nearest neighbor points), the average distance of the nearest neighbors is calculated, and isolated points with a distance greater than "mean + twice the standard deviation" are removed, and then voxel grid downsampling is performed according to "voxel size = component thickness × 1 / 5", and finally the point cloud is translated to the origin of the world coordinate system and scaled to a unit interval to complete the standardization; then, the core features that distinguish the regions are extracted, and the local geometric features are obtained by fitting the local plane with K=20 nearest neighbors to obtain the normal vector. And the curvature is uniformly pointed to the outside of the component, and then obtained by fitting a quadratic surface to obtain the principal curvature. Calculate the Gaussian curvature G= and mean curvature In the planar region, satisfy Global topological features are obtained through Calculate the distance from the point to the reference plane (where...) For: the normal vector of the reference plane, (For reference points on the reference surface); the core segmentation adopts a "region growth + adaptive threshold" strategy, prioritizing the selection of curvature minima or abrupt change points as seed points, and filtering the K nearest neighbors of the seed points according to three rules: the angle between the normal vectors. Gaussian curvature difference (in, Let: the Gaussian curvature of the candidate point to be judged. The criteria are: the Gaussian curvature of the current seed point and the Euclidean distance < twice the average density of the point cloud. If these conditions are met, the seed point is added to the current region and the average feature of the region is updated. This process is repeated until all points are divided. In the post-processing stage, isolated small regions with fewer than 1% of the total number of points are removed, and adjacent regions with an angle between their normal vectors < 1 / 200000 are merged. Curvature difference < The region is then smoothed using moving least squares; finally, a unique identifier is assigned to each sub-region, and labeled point cloud data is output. The system also collects information such as the number of points, average normal vector, and minimum bounding shape for each region, providing data support for subsequent processing.

[0060] In a preferred embodiment of the method, the calculation of the center point coordinates of each partition's 3D point cloud includes:

[0061] The point cloud of partition i is The center point is then calculated as follows: Calculate the distance from each point to the geometric center; points closer to the geometric center have a higher weight. Based on this, recalculate the weighted center. ,in , Let k be the Euclidean distance from the k-th point to the initial geometric center. For scale parameters,

[0062] For each point in the partitioned point cloud, the normal vector is calculated based on its nearest neighbors using principal component analysis: for each point... The set of k nearest neighbors is Calculate the covariance matrix of these points: ,in The mean of the nearest neighbors. For nearest neighbors, with respect to the covariance matrix: Eigenvalue decomposition is performed, and the eigenvector corresponding to the smallest eigenvalue is the surface normal vector at that point. The average or weighted average of the normal vectors of all points within the partition is then calculated to obtain the representative normal vector of that partition. Where N is the number of points in the partition. Let be the normal vector of the j-th point. Finally, normalize the normal vector so that its length is 1.

[0063] In a preferred embodiment of the method, the use of a genetic algorithm to plan the optimal access sequence for each partition includes,

[0064] Using the center point of each partition as a path node, calculate multiple cost functions such as distance between any two partitions, angle between normal vectors, acceleration change, and workload; weight and fuse the cost functions to form a fitness evaluation function; use a genetic algorithm to iteratively search the population for the optimal partition access sequence that minimizes the total movement distance, smooths the turning, and balances the workload through selection, crossover, and mutation operations.

[0065] In a preferred embodiment of the method, the cost function is:

[0066] in The weight coefficients are the partition set. Each partition has a center point. and normal vector Any two partitions and The transfer cost between them is a weighted sum of multiple sub-costs, representing the distance the robotic arm moves: Cost of the angle between normal vectors: Acceleration cost: Workload cost: ;in Let j be the number of points in partition j. To avoid zero denominators for small positive numbers, the cost of maintaining regional continuity is: .

[0067] In a preferred embodiment of the method, camera and robotic arm hand-eye calibration includes:

[0068] The camera is fixed to the end effector fixture of the robotic arm in an eye-out manner, and calibration plate images are acquired in multiple poses during the movement of the robotic arm; the intrinsic parameters of the camera are obtained by Zhang Zhengyou calibration method; the transformation matrix between the camera coordinate system and the robotic arm wrist joint coordinate system is solved by Tsai-Lenz algorithm; laser calibration includes determining the position of the laser source center point in the robotic arm coordinate system and the direction vector of the laser beam.

[0069] In a preferred embodiment of the method, converting the center point coordinates and normal vectors of each partition into the corresponding pose coordinates of the robotic arm includes:

[0070] The coordinates of the center point of the partition under the camera's field of view are transformed to the robot arm's base coordinate system using a hand-eye calibration matrix; the normal vector of the partition is transformed in the corresponding direction using a rotation matrix; the pose coordinates of the robot arm's end effector are calculated based on the transformed center point position and normal vector direction, and the pose coordinates include three-dimensional position coordinates and three Euler angles.

[0071] In a preferred embodiment of the method, ultrafast laser shock peening includes: impacting the surface of an ultrathin component region by region with nanosecond or femtosecond-level ultrafast laser pulses; the laser pulses induce microscopic plastic deformation on the material surface.

[0072] An apparatus for implementing the aforementioned robotic arm-cooperative ultrathin component multi-region reinforcement adaptive path planning method includes,

[0073] A robotic arm system with at least six degrees of freedom to grip a camera and a laser output head;

[0074] A three-dimensional galvanometer scanning system that controls the scanning path of the laser on the surface of a workpiece;

[0075] A camera that provides real-time visual feedback to obtain high-precision 3D point cloud data of the workpiece surface;

[0076] The control system controls the robotic arm system, the three-dimensional galvanometer scanning system 4, the camera, and the laser output head to perform ultrafast laser shock strengthening based on the aforementioned robotic arm-coordinated ultrathin component multi-region strengthening adaptive path planning method.

[0077] In a preferred embodiment of the device, the three-dimensional galvanometer scanning system 4 includes a three-dimensional tunable acousto-optic deflector or dynamic reflector array for three-dimensional scanning of the laser spot position, and an imaging system for focusing the laser onto the workpiece surface.

[0078] In one embodiment, the robotic arm system includes:

[0079] A collaborative robotic arm with at least six degrees of freedom;

[0080] The tooling fixture installed at the end of the robotic arm is used to mount the camera and laser output head;

[0081] Robotic arm controller enables precise control of the robotic arm's movements.

[0082] The binocular line-scanning laser 3D camera includes a dual-line laser light source and a camera sensor array; it is used to acquire high-precision three-dimensional point cloud data of the surface of the ultra-thin component 3 and to provide real-time visual feedback in conjunction with the movement of the robotic arm. Furthermore, the camera is equipped with a camera support 6.

[0083] In one embodiment, according to the appendix Figure 2As shown, the device of this invention comprises the following main components: The robotic arm system uses an industrial collaborative robotic arm with six degrees of freedom, an arm span of not less than 1.2 meters, and a repeatability of not less than ±0.05 mm. A tooling fixture 2 is installed at the end of the robotic arm to simultaneously support a binocular line-scanning laser 3D camera and a laser output head. The robotic arm is driven by a dedicated controller, supporting real-time pose feedback and trajectory planning. The three-dimensional galvanometer scanning system 4 consists of a two-dimensional acousto-optic deflector or a fast-reflecting mirror array, capable of dynamic scanning of the laser position on the XY plane at a frequency of up to 40 kHz. The scanning range is not less than 20 mm × 20 mm at the standard working distance, forming a layered processing mode of "coarse positioning + fine scanning" with the coarse positioning of the robotic arm. The binocular line-scanning laser 3D camera 5 uses an infrared laser source with a wavelength of 1550 nm or visible light, a resolution of not less than 1920 × 1200 pixels, and a frame rate of not less than 8 frames per second. The camera is fixed to the control cabinet 8, maintaining a relatively fixed position and orientation with the laser output head, forming an "eye-outside-hand" configuration. The optical platform 7 and laser 9 utilize ultrafast lasers with pulse widths in the nanosecond or femtosecond range, guiding the laser to the output head at the tooling fixture via optical fiber and an acousto-optic deflector. The optical platform is used to mount the laser, optical path components, and focusing lens, ensuring the stability and focusing effect of the laser optical path 10. The host computer control system runs point cloud data processing software, path planning algorithms, and robotic arm control programs. It communicates in real-time with the robotic arm controller, camera, and laser via Ethernet.

[0084] like Figure 1 As shown, a multi-region adaptive path planning method for strengthening ultrathin components through galvanometer-robotic arm collaboration includes the following steps:

[0085] S1. Ensure the stability of the robotic arm system: Adjust the support feet to keep the robotic arm base in a stable position;

[0086] S2. Power On: Connect to the external power supply line, start the robotic arm 1 and binocular line scanning laser 3D camera 5 through human-machine operation, adjust the exposure, depth range and other parameters of the binocular line scanning laser 3D camera 5 according to the working environment; adjust the parameters of the laser 9 according to the type of ultra-thin component to be processed, and set appropriate speed and other parameters through the robotic arm secondary development kit;

[0087] S3. Ultra-thin component point cloud acquisition: Control the robotic arm to move the component to the initial shooting pose, adjust the exposure, gain and other optical parameters of the binocular line scan laser 3D camera to make the optical field of view of the area to be strengthened reach the ideal acquisition state, and then control the binocular line scan laser 3D camera to scan the surface of the workpiece to be strengthened, obtain the three-dimensional point cloud of the surface of the component to be strengthened and save it. At this time, the pose of the robotic arm wrist joint during shooting should be acquired simultaneously.

[0088] S4. Center Point Coordinate Calculation: The center point coordinates of the 3D point clouds for each region after area division are calculated. After obtaining the 3D point cloud data for each region after area division, the first step in this method is to perform fine feature calculations. The specific process is as follows: First Stage: Data Preprocessing. The collected point clouds of each region are pre-processed. Statistical filtering and radius filtering algorithms are used to denoise the original point clouds, removing outliers introduced by noise from the acquisition system, environmental interference, or data transmission errors. Statistical filtering calculates the average distance from each point to its nearest neighbors and removes points whose distance exceeds a threshold; radius filtering calculates the number of points contained in a spherical neighborhood centered on each point with a set radius. If the number of points is less than the threshold, the point is marked as a noise point. This step ensures the robustness of subsequent calculations. Second Stage: Center Point Calculation. For the point cloud of each region, the geometric center of the point cloud is first calculated. Let the point cloud of region i be... The center point is then calculated as follows: However, to make the center point more representative of the characteristic location of the partition, a weighted center method is further adopted. The distance from each point to the geometric center is calculated, with points closer to the center having a higher weight. Based on this, the weighted center is recalculated. ,in , Let k be the Euclidean distance from the k-th point to the initial geometric center. This is the scale parameter.

[0089] S5. Normal Vector Estimation: For each point in the partitioned point cloud, the normal vector is calculated based on its nearest neighbors. Principal Component Analysis (PCA) is used: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] The set of k nearest neighbors is Calculate the covariance matrix of these points: ,in This is the mean of the nearest neighbors. For Eigenvalue decomposition is performed; the eigenvector corresponding to the smallest eigenvalue is the surface normal vector at that point. The representative normal vector for that partition is obtained by taking the mean or weighted mean of the normal vectors of all points within the partition. Where N is the number of points in the partition. Let be the normal vector of the j-th point. Finally, normalize the normal vector to a length of 1. The output of this stage is the coordinates of the center point of each partition and the unit normal vector; these two key parameters provide the geometric basis for subsequent path planning and coordinate transformation.

[0090] S6. Optimal Access Sequence Planning: After obtaining the center point and normal vector of each partition, a genetic algorithm is used to plan the optimal path and determine the optimal order in which the robotic arm visits each partition. Cost Function Design: This method comprehensively considers multiple factors to construct a cost function. Let the partition set be... Each partition has a center point. and normal vector Any two partitions and The transfer cost between two partition centers is a weighted sum of multiple sub-costs: Distance cost: reflects the Euclidean distance between the two partition centers, representing the distance the robotic arm moves. The cost of the included normal vector angle reflects the degree of difference in the surface orientation of the two partitions. A large included normal vector angle indicates a drastic change in the processing direction of the two partitions, which may require the robotic arm to make significant attitude adjustments. Acceleration cost: Consider the smoothness of the path. If a segment of the path originates from a partition... arrive Then The significant change in direction indicates high acceleration, high energy consumption, and potential impact on the robotic arm. Workload cost: Reflects the processing workload of each partition. Partitions with more points typically require longer processing times; access order should be considered to avoid frequent large-scale switches. ;in Let j be the number of points in partition j. Avoid fractions with small positive numbers where the denominator is zero. The cost of region continuity: prioritize accessing spatially adjacent partitions to reduce large jumps. The overall cost function is: in This is a weighting coefficient, which can be adjusted according to specific working conditions. A typical setting is: .

[0091] This patent models the partition access path planning problem as a variant of the Traveling Salesman Problem (TSP). The specific implementation of the designed genetic algorithm is as follows: 1. Chromosome encoding and population initialization: Each chromosome individual is encoded as a permutation of all partition numbers. Let the partition set be V={1,2,...,n}, then a path can be represented as... ,in and They are all distinct. During initialization, N_pop permutations of this type are randomly generated to form the initial population. The size of N_pop is set according to the problem size (the value of n), for example, N_pop = max(50, 5×n). 2. Fitness Evaluation: The fitness function is directly related to the physical cost of the path. First, define the movement cost between two partitions i and j. (e.g., Euclidean distance or time cost). For a path X, its total cost is: (If a return to the starting point is required) fitness F(X) is set to the reciprocal of the cost, i.e. The lower the cost, the higher the fitness, and the better the individual. 3. Selection operation (roulette wheel selection): Calculate the selection probability of each individual based on the fitness ratio. Individual The probability of being selected is as follows: , by generating N_pop random numbers uniformly distributed in the interval [0, 1), N_pop individuals are selected according to the cumulative probability distribution to form a mating pool, and individuals with higher fitness have a higher probability of being selected multiple times. 4. Crossover operation (Order Crossover OX): Randomly select two parent individuals P1 and P2 from the mating pool. Randomly generate two cutting points a and b (1 ≤ a < b ≤ n). For the generation of offspring C1: a) Copy the segment between cutting points a and b in P1 directly to the same position in C1; b) Starting from the position after the second cutting point of P2, traverse all its partitions in a loop, skip the partitions that have already appeared in the C1 segment, and fill the remaining empty positions in C1 with the remaining partitions in order (starting from position b + 1, wrapping back to position 1 after reaching the end). Generate offspring C2 in the same way with P2 as the segment source. This operation introduces new combinations while retaining the parent sequence segments and naturally ensures the legality of the path (each partition appears once). 5. Mutation operation (Swap Mutation): For each offspring generated by crossover, mutate with probability (usually set to 0.01 - 0.1). If mutation is triggered, randomly select two different position indices u and v in its path and swap the partition numbers at these two positions. For example, before mutation it is (..., A,..., B,...), and after swapping it becomes (..., B,..., A,...). This local perturbation helps the algorithm escape from local optima. 6. Iteration and Convergence: Take the new individual set generated by selection, crossover, and mutation as the next generation population, and repeat steps 2 to 5. The algorithm terminates when it reaches the preset maximum number of iterations (such as 1000 generations) or the improvement amplitude of the optimal fitness in consecutive generations is less than the threshold . Finally, output the partition access sequence corresponding to the individual with the highest fitness in each generation of the population as the optimal path. Compared with the greedy algorithm that only focuses on local optima, this method can systematically search in a larger solution space through population iteration, global crossover exploration, and local mutation perturbation, and thus has a higher probability of obtaining a global approximate optimal solution with a lower total movement cost.

[0092] S7. Hand-eye Calibration: The binocular line-scanning laser 3D camera is calibrated using Zhang Zhengyou's calibration method. This is a crucial step in ensuring the accuracy of visual guidance. Camera Calibration: First, the intrinsic parameters of the binocular line-scanning laser 3D camera are calibrated. A checkerboard calibration board is placed in the camera's field of view, and 20-30 images (n) are taken from multiple different positions and angles. Using Zhang Zhengyou's calibration algorithm, the intrinsic parameter matrix and distortion coefficients of the camera are calculated by the correspondence between the pixel coordinates of key points on the calibration board in the captured images and the world coordinates. These parameters are used to subsequently convert the pixel coordinates into three-dimensional coordinates in the camera coordinate system. Hand-eye Calibration: With the eye outside the hand, the camera is fixed to the end effector fixture of the robotic arm. Let the base coordinate system of the robotic arm be Base, the coordinate system of the wrist joint end flange be Wrist, and the camera coordinate system be Camera. The goal of hand-eye calibration is to solve for the rigid body transformation matrix from Wrist to Camera. The specific method is as follows: the robotic arm moves to different positions and postures according to a preset trajectory, and the position and posture of the robotic arm's wrist joint are recorded at each position. (The k-th position). Simultaneously, the camera captures an image of the calibration plate and calculates its position in the camera coordinate system. According to the chain rule of rigid body transformation: If the calibration plate remains fixed in the base coordinate system, multiple sets of equations can be established and solved using the Tsai-Lenz algorithm or nonlinear optimization methods. After calibration, verify the calibration accuracy. Select several new robotic arm poses, perform coordinate transformation using the calibration results, and compare the deviation between the predicted and actual positions. If the deviation is within the allowable range of 0.1 mm, the calibration is successful; otherwise, the calibration needs to be repeated.

[0093] S8. Laser Calibration: The purpose of laser calibration is to determine the spatial position of the laser output head and the direction of the laser beam. Since the laser is connected to the end effector of the robotic arm via a fiber optic coupler, and the position of the laser output head is fixed on the same fixture as the camera, their relative position needs to be determined through calibration. The method is as follows: On a calibration target visible to both the camera and the laser output head, record the position of the laser spot under different robotic arm poses. Use the camera to perform high-precision detection of the spot position, establishing the relationship between the spot pixel coordinates and the robotic arm pose. Using a 3D perspective method (PnP) or a similar calibration algorithm, the position of the laser output head relative to the camera coordinate system can be calculated. and the direction vector of the laser beam .

[0094] S9. Center Point Coordinate Transformation: Based on the eye-in-hand visual guidance theory, the center point and normal vector of a partition are transformed from the camera coordinate system to the robot arm's base coordinate system, thereby calculating the robot arm's target pose. Center Point Coordinate Transformation: Let the homogeneous coordinates of the center point of a certain partition in the camera coordinate system be... The transformation matrix obtained through hand-eye calibration. and the current wrist joint pose of the robotic arm The position of this point in the base coordinate system can be calculated: Matrix multiplication is performed according to standard linear algebra rules. The resulting transformation gives the three-dimensional coordinates of the partition's center point in the robot arm's base coordinate system. ;

[0095] S10. Normal Vector Direction Transformation: The normal vector is a direction quantity and does not involve translation; it only requires transformation using a rotation matrix. Let the partition normal vector in the camera coordinate system be... ,extract The rotating part is obtained The normal vector is transformed into: Further, extract rotation matrix This gives the direction of the normal vector in the base coordinate system: ;

[0096] S11. Target Pose Calculation: The target pose of the robotic arm consists of two parts: target position and target orientation. The target position has been obtained by transforming the coordinates of the center point. The target orientation needs to be determined according to the machining process. For laser enhancement applications, the laser should be perpendicular to the workpiece surface, i.e., along the normal vector direction of the surface, to obtain the best effect. Therefore, the target orientation should ensure that the downward direction of the robotic arm's end effector is consistent with the normal vector direction of the partition. Assuming that the initial forward direction of the current end effector is downward (-Z direction), it needs to be rotated to the normal vector direction. Construct the rotation matrix. This makes its third column equal to the transformed unit normal vector. Euler angle decomposition yields three Euler angles (rotation angles about the X, Y, and Z axes). The target pose coordinates are represented as a six-dimensional vector: The first three components are positions, and the last three components are Euler angles;

[0097] S12. Perform laser strengthening processing: Control the robotic arm to begin laser strengthening according to the planned path. Trajectory execution: Based on the optimal partition access sequence output by the genetic algorithm (e.g., ... The host computer sequentially reads the target pose of each partition. The pose commands are sent to the robotic arm controller via industrial Ethernet or other real-time communication protocols. Based on these commands, the robotic arm controller calculates the target angles of each joint through inverse kinematics and controls the joint motors to smoothly move to the specified position and posture via a servo drive system. After the robotic arm moves to the target position of a certain partition, its position is fine-tuned. Real-time visual feedback is provided using a camera; if there is a deviation between the actual position and the target position, a correction amount is calculated based on the feedback information, and a fine-tuning command is issued until the required positioning accuracy is achieved.

[0098] Galvanometer Scanning Control: After the robotic arm is positioned at the partition location, the 3D galvanometer system begins operation. The host computer calculates the two-dimensional trajectory that the galvanometer needs to scan based on the geometric information of the partition. For rectangular or square partitions, it can scan straight lines or grid patterns; for complex curved surface partitions, it can calculate an adaptive scanning path to ensure uniform coverage. The galvanometer, according to the issued scanning command, rapidly changes the laser direction at a high frequency, causing the laser spot to scan the partition surface along a predetermined trajectory. Simultaneously, pulse parameters such as laser pulse energy and repetition frequency can also be dynamically adjusted.

[0099] Process Monitoring and Feedback: During scanning, the system continuously monitors key parameters such as laser output power and robotic arm position drift. If an anomaly is detected, the control system immediately issues an alarm and suspends processing to prevent workpiece quality issues. After processing is complete, the camera re-images that section of the workpiece to obtain a post-processed point cloud, which is compared with the pre-processed point cloud. The processing effect is evaluated through factors such as surface roughness and residual stress distribution. If the evaluation results meet the process requirements, processing continues to the next section; otherwise, reprocessing or adjustment of process parameters may be necessary.

[0100] Furthermore, the intelligent partitioning feature calculation based on point cloud in this invention not only achieves high-fidelity extraction of complex surface geometry information of ultra-thin components, but also significantly improves the representativeness and robustness of partition center points and surface normal vectors by combining weighted distance search with principal component analysis (PCA). This technique provides accurate geometric basis for subsequent robotic arm pose generation, effectively avoiding laser incident angle deviation caused by feature point deviation, thereby ensuring the consistency of laser shock strengthening effect and process stability.

[0101] The multi-objective path optimization mechanism based on genetic algorithms breaks through the limitations of traditional fixed-sequence or locally optimal path planning, modeling the partition access problem as a comprehensive optimization problem integrating spatial distance, posture change, motion smoothness, load balancing, and regional continuity. This mechanism significantly reduces the ineffective large-scale movements of the robotic arm during multi-region switching, lowers joint acceleration fluctuations and energy consumption, and improves the overall processing cycle time and equipment utilization rate, making it particularly suitable for typical application scenarios such as thin-walled aerospace components with multiple discrete reinforcement regions. The high-precision coordinate transformation system guided by hand-eye vision constructs a closed-loop mapping relationship from visual perception to mechanical execution through a rigorous triple calibration process of camera intrinsic parameter calibration, hand-eye calibration, and laser optical axis calibration. This system ensures that the processing features extracted from point cloud data can be accurately mapped to the robotic arm's base coordinate system and further transformed into the target pose of the end effector, fundamentally solving the coordinate consistency problem between "seeing" and "accurately hitting" in non-contact laser processing, and meeting the stringent requirements of micron-level positioning accuracy for ultra-thin components. The galvanometer-robotic arm hierarchical collaborative control architecture achieves a highly efficient "coarse positioning + fine scanning" operation mode: the six-axis robotic arm is responsible for rapidly delivering each enhancement area into the effective processing field of view of the galvanometer, while the three-dimensional galvanometer dynamically adjusts the laser focus position at kHz-level frequencies within a tiny area to complete a high-density, high-uniformity scanning trajectory. This tight coupling between the two in the spatiotemporal dimension leverages both the robotic arm's wide-range flexibility and the galvanometer's high-speed, high-precision fine-tuning capabilities, thereby significantly improving processing efficiency while ensuring processing quality.

[0102] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0103] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for multi-region reinforcement adaptive path planning of ultra-thin components in cooperation with a robot arm, characterized in that, It includes, A binocular line-scanning laser 3D camera was used to acquire three-dimensional data of the surface of the ultrathin component, and a region division algorithm was used to divide the three-dimensional point cloud of the ultrathin component surface into regions. The center point coordinates and normal vectors of the 3D point clouds of each region after regional division are calculated and estimated. A genetic algorithm is used to plan the optimal access sequence for each partition and determine the optimal processing order. Complete the calibration of the camera and robotic arm's hand-eye alignment, as well as the calibration of the laser center point and laser direction of origin; Convert the center point coordinates and normal vectors of each partition into the corresponding pose coordinates of the robotic arm; The robotic arm is controlled to place each section of the ultrathin component sequentially into the processing area of ​​the three-dimensional galvanometer according to the planned optimal access sequence and the pose coordinates. The galvanometer then performs ultrafast laser shock blasting on the surface of the component.

2. The method as described in claim 1, characterized in that, Preferably, the calculation of center point coordinates and estimation of normal vectors for the 3D point clouds of each region after regional division includes, Extract the 3D point cloud data of each partition; Calculate the geometric center of the point cloud for each partition as the coordinates of the center point of that partition; Principal component analysis of the covariance matrix of nearest neighbors is used to calculate the normal vector of the partition surface, and then the normal vector is normalized.

3. The method as described in claim 2, characterized in that, The calculation of the center point coordinates for each partition's 3D point cloud includes: The point cloud of partition i is The center point is then calculated as follows: Calculate the distance from each point to the geometric center; points closer to the geometric center have a higher weight. Based on this, recalculate the weighted center. ,in , Let k be the Euclidean distance from the k-th point to the initial geometric center. For scale parameters; For each point in the partitioned point cloud, the normal vector is calculated based on its nearest neighbors using principal component analysis: for each point... The set of k nearest neighbors is Calculate the covariance matrix of these points: ,in The mean of the nearest neighbors. For nearest neighbors, with respect to the covariance matrix: Eigenvalue decomposition is performed, and the eigenvector corresponding to the smallest eigenvalue is the surface normal vector at that point. The average or weighted average of the normal vectors of all points within the partition is then calculated to obtain the representative normal vector of that partition. Where N is the number of points in the partition. Let be the normal vector of the j-th point. Finally, normalize the normal vector so that its length is 1.

4. The method as described in claim 1, characterized in that, The optimal access sequence planning for each partition is performed using a genetic algorithm, including... Using the center point of each partition as a path node, calculate multiple cost functions such as distance between any two partitions, angle between normal vectors, acceleration change, and workload; weight and fuse the cost functions to form a fitness evaluation function; use a genetic algorithm to iteratively search the population for the optimal partition access sequence that minimizes the total movement distance, smooths the turning, and balances the workload through selection, crossover, and mutation operations.

5. The method as described in claim 4, characterized in that, The cost function is: in The weight coefficients are the partition set. Each partition has a center point. and normal vector Any two partitions and The transfer cost between them is a weighted sum of multiple sub-costs, representing the distance the robotic arm moves: Cost of the angle between normal vectors: Acceleration cost: Workload cost: ;in Let j be the number of points in partition j. To avoid zero denominators for small positive numbers, the cost of maintaining regional continuity is: .

6. The method as described in claim 1, characterized in that, Camera and robotic arm hand-eye calibration includes, The camera is fixed to the end-effector of the robotic arm in an eye-out manner, and the calibration plate images are acquired in multiple poses during the movement of the robotic arm. The camera's intrinsic parameters were obtained using the Zhang Zhengyou calibration method; the transformation matrix between the camera coordinate system and the robotic arm wrist joint coordinate system was solved using the Tsai-Lenz algorithm; laser calibration included determining the position of the laser source center point in the robotic arm coordinate system and the direction vector of the laser beam.

7. The method as described in claim 1, characterized in that, The center point coordinates and normal vectors of each partition are converted into the corresponding pose coordinates of the robotic arm, including: The coordinates of the center point of the partition under the camera's field of view are transformed to the robot arm's base coordinate system using a hand-eye calibration matrix; The normal vectors of the partition are transformed into the corresponding directions using a rotation matrix; the pose coordinates of the robotic arm end effector are calculated based on the transformed center point position and normal vector direction. The pose coordinates include three-dimensional position coordinates and three Euler angles.

8. The method as described in claim 1, characterized in that, Ultrafast laser shock peening includes: impacting the surface of ultrathin components region by region with nanosecond or femtosecond ultrafast laser pulses; and inducing microscopic plastic deformation on the material surface by laser pulses.

9. An apparatus for implementing the robotic arm-assisted multi-region reinforcement adaptive path planning method for ultrathin components as described in any one of claims 1-8, characterized in that, It includes, A robotic arm system with at least six degrees of freedom to grip a camera and a laser output head; A three-dimensional galvanometer scanning system that controls the scanning path of the laser on the surface of a workpiece; A camera that provides real-time visual feedback to obtain high-precision 3D point cloud data of the workpiece surface; The control system controls the robotic arm system, the three-dimensional galvanometer scanning system, the camera, and the laser output head to perform ultrafast laser shock strengthening based on the aforementioned robotic arm-coordinated ultrathin component multi-region strengthening adaptive path planning method.

10. The apparatus as claimed in claim 9, characterized in that, The three-dimensional galvanometer scanning system includes a three-dimensional tunable acousto-optic deflector or dynamic mirror array for three-dimensional scanning of the laser spot position, and an imaging system for focusing the laser onto the workpiece surface.

Images