A method and system for automatic planning of laser surface machining path of a profiled member

Abstract

The application provides a kind of special-shaped component laser surface processing path automatic planning method and system, belong to robot laser surface processing path planning technical field, the method: two postures are respectively collected two frames of three-dimensional point cloud data of special-shaped component surface and are preprocessed;Two frames of three-dimensional point clouds after pre-processing are registered, obtain complete component surface topography point cloud and establish local coordinate system;The point cloud after registration is input into the pre-trained three-dimensional depth completion neural network, and a high-fidelity three-dimensional topography model is reconstructed;The feature point cloud of the region to be processed is extracted by inputting the reconstructed model into the improved region growing segmentation algorithm;Determine the slice parameters according to the feature point cloud, fit the spline curve and discretize to obtain the robot path point;The robot moves along the path, and cooperates with the laser to complete full coverage processing.The application does not need to rely on CAD model and manual teaching, and realizes the automatic planning of laser surface processing path of complex curved surface special-shaped component.

Description

Technical Field

[0001] This application belongs to the field of robot laser surface processing path planning technology, specifically relating to an automatic planning method and system for laser surface processing paths of irregularly shaped components. Background Technology

[0002] Laser surface processing is an advanced manufacturing technology that uses the interaction between a laser beam and the surface of a material to strengthen, modify, repair, or precisely process the surface of a component. It features controllable energy, non-contact operation, a small heat-affected zone, and high processing flexibility.

[0003] In modern aerospace, energy equipment, and high-end manufacturing, many components possess irregular structural features such as thin walls, curved surfaces, variable cross-sections, or complex local characteristics. During service, these components typically endure complex conditions including airflow erosion, vibration loads, thermal cycling, and contact wear. Their surface quality and local properties directly impact their reliability and service life. Therefore, highly consistent laser processing path planning for irregularly shaped component surfaces is a crucial foundation for achieving automated and precise surface treatment.

[0004] With the development of intelligent manufacturing and robotic processing technologies, laser surface processing systems using industrial six-axis robots as the main motion source are attracting increasing attention. Path planning for robotic laser surface processing systems is a crucial preparatory task that must be completed before processing. Currently, laser processing paths for complex curved surfaces mainly rely on manual teaching or offline programming. Manual teaching is inefficient and cannot meet the need for rapid generation of enhanced paths for complex curved surfaces; offline programming relies on accurate prior CAD models and stable clamping postures, making it difficult to adapt to situations where there are no CAD models or where the actual surface has geometric errors and posture deviations. Existing research has utilized computer graphics to achieve region division or automatically generated paths through point cloud technology, but methods for laser surface processing path planning of irregularly shaped components based on 3D point clouds are still relatively scarce, especially lacking automatic path generation methods for complex curved surfaces that can achieve path generation without manual teaching.

[0005] In summary, there is an urgent need for an automatic laser surface processing path planning method that does not rely on CAD prior models and can achieve manual teaching-free processing of complex curved and irregular components, so as to simultaneously meet the requirements of coverage consistency, normal incidence, and processing efficiency without accurate geometric models. Summary of the Invention

[0006] In a first aspect, embodiments of this application provide an automatic planning method for laser surface processing paths of irregularly shaped components, comprising the following steps: S1. Acquire two frames of 3D point cloud data of the surface of the irregular component in two different postures, and preprocess each frame of 3D point cloud data separately; S2. Input the two preprocessed frames of 3D point cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; S3. Input the registered complete component surface topography point cloud into the pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. S4. Input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. S5. Determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, fit the spline curve based on the sliced ​​point cloud, and discretize the spline curve to obtain the robot path points. S6. Control the robot to move along the path according to the robot path points, and cooperate with the laser to perform laser shock strengthening processing until the full coverage processing is completed.

[0007] Secondly, embodiments of this application also provide an automatic planning system for laser surface processing paths of irregularly shaped components, comprising: The point cloud acquisition module is used to acquire two frames of three-dimensional point cloud data from the surface of the irregular component in two different orientations, and to preprocess each frame of three-dimensional point cloud data separately. The registration module is used to input the preprocessed two frames of 3D point cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; The depth completion module is used to input the registered complete component surface topography point cloud into a pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. The region segmentation module is used to input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. The path planning module is used to determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, to slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, to fit the spline curve based on the sliced ​​point cloud, and to discretize the spline curve to obtain the robot path points. The robot control module is used to control the robot to move along the path according to the robot's path points, and to cooperate with the laser to perform laser shock enhancement processing until the full coverage processing is completed.

[0008] As can be seen from the above technical solutions, this application has the following advantages: The automatic planning method and system for laser surface processing of irregularly shaped components provided in this application achieves automatic planning of laser surface processing paths for complex curved irregularly shaped components without the need for CAD models or manual teaching; it achieves high-fidelity 3D reconstruction through PENet depth completion, solving the data loss problem caused by reflective occlusion; it accurately extracts the processing area through improved region growth segmentation using curvature sorting and dual threshold judgment; and it combines overlap rate and spot diameter to constrain slice spacing, and ensures path smoothness and normal incidence through B-spline fitting and normal vector calculation, ensuring uniform processing coverage and consistent strengthening effect. Attached Figure Description

[0009] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the automatic planning method for laser surface processing paths of irregularly shaped components according to the present invention.

[0011] Figure 2 This is a schematic diagram of the automatic planning system for laser surface processing paths of irregularly shaped components according to the present invention.

[0012] Figure 3 This is a flowchart of the Cauchy-weighted point-to-surface ICP algorithm in this invention.

[0013] Figure 4 This is a flowchart of the PENet deep completion algorithm in this invention.

[0014] Figure 5 This is a schematic diagram of obtaining slice data based on the strip-shaped bilateral nearest neighbor method in this invention.

[0015] Figure 6 This is a flowchart of the robot path point acquisition algorithm in this invention.

[0016] Figure 7 This is a schematic diagram of the laser surface processing system for irregularly shaped components in this invention.

[0017] Among them, 1-industrial robot; 2-irregularly shaped component; 3-structured light sensor; 4-laser. Detailed Implementation

[0018] The various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the automatic planning method for laser surface processing of irregularly shaped components. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0019] This embodiment provides an automatic planning method for laser surface processing paths of irregularly shaped components. It eliminates the need for CAD models and manual teaching, enabling automatic planning of laser surface processing paths for complex curved irregularly shaped components and ensuring consistent processing coverage and normal incidence.

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 The diagram shows a flowchart of an automatic path planning method for laser surface processing of irregularly shaped components in a specific embodiment. The method includes the following steps: S1. Acquire two frames of 3D point cloud data of the surface of the irregular component in two different postures, and preprocess each frame of 3D point cloud data separately; It should be noted that by acquiring two frames of 3D point cloud data of the surface of the irregular component in two poses, the problem of data occlusion caused by the complex curved surface of the component under a single viewpoint is overcome; and outliers and redundant data are effectively removed through preprocessing, which improves the efficiency and accuracy of registration. S2. Input the two preprocessed frames of 3D point cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; It should be noted that by using a registration algorithm to fuse two frames of 3D point clouds into a complete surface morphology, the problem of limited field of view and incomplete information of a single frame of 3D point cloud is solved. By establishing a local coordinate system aligned with the Z-axis of the robot's end flange, a unified coordinate reference is established for path planning and robot control, ensuring the accuracy of coordinate transformation during the processing. S3. Input the registered complete component surface topography point cloud into the pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. It should be noted that by using a pre-trained neural network and the texture information of RGB images to help repair the missing points in the point cloud, the problem that scanners cannot acquire data due to reflection and transmission of irregularly shaped components is solved, and a dense point cloud model that can be used for accurate calculation is generated. S4. Input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. It should be noted that traditional region growing is prone to overflow at the edges or interference from noise. This step uses curvature sorting to select seeds and dual threshold determination of normal and curvature, which can accurately separate the irregular region to be processed from the complex background or workpiece fixture, avoiding misprocessing. S5. Determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, fit the spline curve based on the sliced ​​point cloud, and discretize the spline curve to obtain the robot path points. It should be noted that the overlap rate of the laser spot is guaranteed by adaptively determining the slice spacing by calculating the point cloud density, the smoothness of the robot motion is guaranteed by B-spline fitting, and discretization optimizes the amount of data, thus transforming the disordered point cloud into ordered path points. S6. Control the robot to move along the path according to the robot path points, and cooperate with the laser to perform laser shock strengthening processing until the full coverage processing is completed; It should be noted that the path points are converted into the pose of the robot's end effector, and the laser is controlled to synchronize with the robot to ensure that the laser beam is always incident perpendicularly on the workpiece surface, thereby achieving full coverage and uniform strengthening.

[0022] This embodiment eliminates the need for CAD models and manual teaching, enabling automatic planning of laser surface processing paths for complex curved and irregularly shaped components. High-fidelity 3D reconstruction is achieved through a depth completion network, and the processing area is accurately extracted through improved region growth segmentation. A smooth path is generated by combining process constraint slices and B-spline fitting, ensuring consistent coverage and normal incidence.

[0023] In this embodiment, irregularly shaped components refer to components with irregular geometric structures such as thin walls, curved surfaces, variable cross-sections, or complex local features, and their surfaces are complex free-form surfaces. The structured light sensor uses a binocular structured light camera with a resolution of no less than 1280×1024 pixels and a measurement accuracy better than 0.05mm. The industrial robot is a six-axis serial industrial robot with a repeatability accuracy better than ±0.05mm. The laser is a nanosecond pulsed laser with a wavelength of 1064nm, a pulse width of 10ns-30ns, and a single pulse energy of 2J-10J.

[0024] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process in this embodiment, another automatic planning method for laser surface processing paths of irregularly shaped components is provided. This method includes the following steps: S1. Acquire two frames of 3D point cloud data of the surface of the irregular component in two different postures, and preprocess each frame of 3D point cloud data separately; The specific steps of step S1 are as follows: S11. Secure the irregularly shaped component to the end flange of the industrial robot; S12. Position the irregular component in the first posture directly in front of the structured light sensor and collect the first frame of three-dimensional point cloud data on the surface of the irregular component. S13. Position the irregular component in the second posture directly in front of the structured light sensor and collect the second frame of three-dimensional point cloud data on the surface of the irregular component. S14. Perform radius filtering, noise reduction and downsampling operations on the first frame of 3D point cloud data and the second frame of 3D point cloud data respectively to obtain the first preprocessed point cloud and the second preprocessed point cloud. It should be further explained that in step S14, radius filtering is used to remove isolated outliers in the point cloud. The filtering radius is 2-3 times the average spacing of the point cloud, and the threshold for the number of neighboring points is 10-20. Statistical filtering is used for denoising. The statistical distribution of the neighborhood distance of each point is analyzed, and points whose mean exceeds 2 standard deviations of the global mean are removed. Voxel filtering is used for downsampling. The voxel side length is selected based on the balance between subsequent registration accuracy and computational efficiency, preferably 0.2mm-0.5mm. S2. Input the preprocessed two frames of 3D cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; The specific steps of step S2 are as follows: S21. Use the first preprocessed point cloud as the source point cloud and the second preprocessed point cloud as the target point cloud; S22. Source point cloud Each point in the target point cloud The algorithm uses a KD-Tree search to find the nearest point, establishes the correspondence between initial points, and sets the maximum number of iterations, the convergence threshold, and the Cauchy scale parameter. The initial value; It should be further explained that in step S22, the scale parameter c controls the weighting intensity of the Cauchy weighting function, and its value is adaptively adjusted according to the noise level of the point cloud data. ,in, The value of c is estimated by calculating the standard deviation of the residuals from each point in the point cloud to its neighborhood fitting plane. When the noise level is high, the value of c is increased, so that more points get greater weight and avoid over-suppressing the true corresponding points. When the noise level is low, the value of c is decreased, which enhances the suppression effect on external points. S23. Based on the current correspondence, construct the distance error function from the point to the surface, and use the Levenberg-Marquardt algorithm to iteratively optimize the transformation matrix. Minimize the following weighted objective function:

[0025] in, Let be a rotation matrix. It is a translation vector. Source Point Cloud The i-th point in For target point cloud Zhongyu The corresponding nearest point; for The normal vector at point N; N is the number of points in the source point cloud; The Cauchy weighting function is expressed as follows:

[0026] in, Let be the point-to-surface distance error of the i-th pair of corresponding points. To control the scale parameters of the weighting intensity; Scale parameters The noise level is adaptively adjusted based on the noise level of the point cloud data, which is estimated by the average distance from each point in the point cloud to its neighborhood fitting plane. like Figure 3 As shown, the Cauchy-weighted point-to-surface ICP algorithm flow is as follows: First, input the source point cloud and the target point cloud, and set the initial transformation matrix, convergence threshold, maximum number of iterations, and initial value of the Cauchy scale parameter c; then, for each point in the source point cloud, search for the nearest point in the target point cloud to establish a correspondence; based on the current correspondence, construct the point-to-surface distance error function, and use the Levenberg-Marquardt algorithm to iteratively optimize the transformation matrix. Minimize the weighted objective function; apply the optimized transformation matrix to the source point cloud and update its position; determine whether the change in the transformation matrix is ​​less than the convergence threshold or the maximum number of iterations has been reached. If so, output the final transformation matrix; otherwise, re-search for the nearest point correspondence and continue iterating. S24. Apply the current optimized transformation matrix. Effect on source cloud Update the source point cloud location; S25. Search for the nearest point correspondence between the updated source point cloud and the target point cloud again, and repeat steps S23 to S24 for iterative optimization; S26. Stop iterating when the change in the transformation matrix is ​​less than the convergence threshold or the maximum number of iterations is reached, and output the final transformation matrix. ; S27. Transform the final matrix Effect on source cloud The source point cloud is transformed to the target point cloud coordinate system to obtain a complete point cloud of the component surface morphology. ; S28. Point cloud of component surface morphology Principal component analysis is used to calculate three principal component directions. A local coordinate system is established, aligned with the Z-axis of the robot's end effector flange, with the first principal component direction as the X-axis, the second principal component direction as the Y-axis, and the third principal component direction as the Z-axis. The complete point cloud with this local coordinate system is then output. ; S3. Input the registered complete component surface topography point cloud into the pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. The specific steps of step S3 are as follows: S31. Use an RGB camera to acquire RGB images of the surface of the irregularly shaped component. ; S32. Transform the complete point cloud with a local coordinate system. With RGB images As input data, it is fed into a pre-trained PENet deep completion model; The PENet deep completion model includes an encoder and a decoder. The encoder processes RGB images... and complete point cloud Feature extraction is performed separately, and the decoder fuses the extracted multimodal features to obtain a complete point cloud. The system performs point-by-point completion of depth information missing due to reflections, occlusions, or sensor limitations, and outputs a completed dense point cloud. ; like Figure 4As shown, the algorithm flow of the PENet depth completion model is as follows: First, the registered complete point cloud and RGB image are used as input. The encoder of the PENet model contains two parallel feature extraction branches. The RGB image branch uses a pre-trained ResNet-34 backbone network to extract multi-scale visual features, and the point cloud branch uses PointNet++ to extract geometric features. The outputs of the two branches are aligned and fused through an adaptive attention fusion module. The decoder uses a 3D deconvolutional network based on sparse convolution. It takes the fused features as input, upsamples stepwise, and outputs a dense point cloud to complete the depth information missing due to reflection, occlusion, or sensor limitations point by point. S33. Dense point cloud Output as a high-fidelity 3D topography reconstruction model; It should be further explained that the PENet deep completion model adopts an encoder-decoder structure. The encoder consists of two parallel feature extraction branches: the RGB image branch uses a pre-trained ResNet-34 backbone network to extract multi-scale visual features, and the point cloud branch uses PointNet++ to extract geometric features. The outputs of the two branches are aligned and fused through an adaptive attention fusion module. The decoder uses a 3D deconvolutional network based on sparse convolution, which takes the fused features as input, progressively upsamples, and outputs a dense point cloud. The PENet deep completion model was pre-trained on the NYU-Depth-v2 public dataset, and the pre-training loss function is composed of a weighted sum of Chamfer Distance loss and Huber loss. S4. Input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. The specific steps of step S4 are as follows: S41. High-fidelity 3D topography reconstruction model Each point in Take the K nearest neighbors in the local neighborhood to form the neighborhood point set. Calculate the neighborhood point set covariance matrix :

[0027] in, For neighborhood point set The j-th point in For neighborhood point set The center of mass, ; For covariance matrix Eigenvalue decomposition yields three eigenvalues. Arranged in ascending order as follows The corresponding feature vectors are respectively ,satisfy , It is the index of the eigenvalue. ; Take the smallest eigenvalue Corresponding feature vector As a point normal vector ,Right now ; The point is defined as the proportion of the smallest eigenvalue to the sum of all eigenvalues. Local curvature ,Right now ; Spatial coordinates of the junction Generate a feature vector for each point ; S42. High-fidelity 3D topography reconstruction model In the process, all undivided points are used to form a candidate point set, and the curvature of each point is considered. Sort the values ​​in ascending order, select the undivided point with the smallest curvature value as the initial seed point, and add the initial seed point to the seed list. ; S43. From the seed list Select one point from the list as the current seed point. For the current seed point The KD-Tree search is used to find its K nearest neighbors in space, forming a nearest neighbor set. ; S44. For the neighbor set Each neighboring point in The following three constraints will be evaluated sequentially: Spatial distance constraints: Neighboring points To the current seed point European distance Less than the preset first distance threshold ; in, The calculation method is as follows:

[0028] Take neighboring points To the current seed point The median of the Euclidean distance; Normal vector angle constraint condition: Neighboring points With the current seed point The angle between the normal vectors Less than the preset angle threshold , The calculation method is as follows:

[0029] In the formula, Neighboring points The normal vector, The current seed point The normal vector; Curvature constraint conditions: Neighboring points curvature Less than the preset curvature threshold ; If all three conditions are met, then the nearest points will be... Add to current area ; In step S44, neighboring points It also needs to satisfy the orthogonal distance constraint to the fitting plane, i.e., the nearest point orthogonal distance to its local fitting plane Less than the preset second distance threshold ; orthogonal distance The calculation method is as follows:

[0030] in, Neighboring points The centroid of the local neighborhood in the plane, where n is the plane normal vector; Second distance threshold The calculation method is as follows:

[0031] in, The median of the orthogonal distances between all points in the nearest neighbor set. The median error of the orthogonal distance is calculated as follows:

[0032] In the formula, Let be the orthogonal distance from the nearest neighbor point to the fitted plane. This indicates that the median is taken, and 1.4826 is the consistency constant for converting the absolute deviation of the median to the standard deviation scale; The constant 1.4826 comes from the quantile function Φ of the standard normal distribution. - The reciprocal of ¹(3 / 4)≈0.6745, i.e., 1 / Φ -¹(3 / 4)≈1.4826 is used to keep the absolute deviation of the median consistent with the standard deviation on scale, thereby making the threshold estimate robust to outliers; S45. Add the new item to the current area. The point is added to the seed list as a new seed point. and the currently processed seed point From the seed list Remove from; S46. Repeat steps S43 to S44 until the seed list is complete. Empty, current region Growth complete, and the current area As a clustering region Output; S47. High-fidelity 3D topography reconstruction model Remove the already segmented points, return to step S42, and repeat until all points are segmented, resulting in multiple clustered regions. ; S48. Based on the preset minimum clustering threshold and maximum number of points threshold Each cluster region was filtered, and those with fewer than [number of points] were removed. or more Within the specified region, clustered regions that meet the threshold range are retained and merged to form the feature point cloud of the region to be processed. ; It should be further explained that in step S41, the value of K is adaptively adjusted according to the point cloud density, with an optimal range of 20-50; when the point cloud density is high, K takes a smaller value to ensure the detail resolution of local features; when the point cloud density is low, K takes a larger value to ensure the statistical stability of the covariance matrix; in step S44, a preset angle threshold is used. The value range is 5°-15°, and the curvature threshold is... The value range is 0.01-0.05, and the above threshold can be adjusted according to the geometric features of the area to be processed and the processing accuracy requirements; In step S44, a triple determination is made using spatial distance constraints, normal vector angle constraints, and curvature constraints to ensure that only points with highly similar geometric features are grouped into the same cluster region, avoiding the edge overflow problem that traditional region growing algorithms are prone to in geometrically abrupt or noisy regions; in step S48, the minimum number of cluster points threshold is set. and maximum number of points threshold Based on the point cloud density and the area of ​​the region to be processed, it is used to eliminate false small clusters caused by noise and large clusters caused by undersegmentation; S5. Determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, fit the spline curve based on the sliced ​​point cloud, and discretize the spline curve to obtain the robot path points. The specific steps of step S5 are as follows: S51. Calculate the feature point cloud of the region to be processed. Average point cloud density Specifically: For the feature point cloud of the region to be processed with N points ,use The Euclidean distance between points p and q is expressed by... This represents the distance between point p and its nearest neighbor: ,

[0033] Then feature point cloud Average point cloud density for:

[0034] S52. Based on average point cloud density Overlap rate and spot diameter Determine the point cloud slicing parameters, including the slice spacing. With slice thickness ; Slice spacing Determined by the spot diameter and overlap ratio: ; slice thickness Based on average point cloud density Set as ; in This is the proportionality coefficient; S53. Construct slice data based on the strip-shaped bilaterally nearest neighbor method to obtain the slice data point set of all layers. Where M is the number of slice layers, This represents the slice data for the m-th layer; In step S53, the specific steps are as follows: S531. Define the slicing plane E as the XY plane, and establish a local coordinate system by taking the normal vector of the slicing plane E as the Z-axis; S532. Feature point cloud to be sliced As input, points located in the positive half-space of the slice plane E are assigned to the positive point set. ,set up For the point set The point in the middle, ; Points located in the negative half-space of the slice plane E are classified into the negative point set. ,set up negative point set The point in the middle, ; S533. Point Set and negative point set Construct KD-Tree data structures along the X and Y directions respectively to obtain positive data structures. and negative data structures ; S534. For the point set Each point in In negative data structures The nearest point in the search space So that the two points satisfy:

[0035] S535. Judgment Point With point Is the Euclidean distance less than the specified search radius r? If not, discard the current point and process the set of points. The next point in; If so, proceed to S536; S536. Connection Point and points Given a line segment L, find the intersection point of line segment L and the slicing plane E. : Set the intersection point Located on line segment L, let the parameter t represents the proportion of the intersection point on line segment L (when t=0, it is located at point L). At t=1, it is located at At point t ∈ [0, 1] and When = 0, the intersection point lies on the slice plane. From the parametric equation of the spatial line, the coordinates of the intersection point can be obtained as:

[0036] S537. Correct Point Set Repeat steps S534 to S536 for each point to obtain all intersection points under the current slice plane E, which constitute the slice data point set of this layer. ; S538. Along the Z-axis, according to the slice spacing Translate slice plane E, repeat steps S532 to S537, and traverse the entire feature point cloud. Obtain the slice data point set of all layers. , where M is the number of slice layers; like Figure 5 As shown, the principle of constructing slice data based on the strip-shaped bilateral nearest neighbor method (i.e., intersection method) is as follows: Define the slice plane E as the XY plane, and take the normal vector of the slice plane E as the Z-axis to establish a local coordinate system; assign points located in the positive half space of the slice plane E to the positive point set. Points located in the negative half-space of the slice plane E are assigned to the negative point set. ;right Each point in ,exist The nearest point in the search space ,connect and Given a line segment L, find the intersection point of line segment L and the slicing plane E. ; through the Repeat the above operation for each point to obtain all intersection points under the current slice plane, which constitute the slice data point set of this layer; translate the slice plane E along the Z-axis according to the slice spacing, repeat the above process, traverse the entire feature point cloud, and obtain the slice data of all layers. S54. For each layer of sliced ​​data Perform uniform downsampling to reduce the amount of data while maintaining the slice shape, resulting in the downsampled slice point set. ; S55. For each downsampled slice point set Smooth curve fitting is performed using cubic quasi-uniform B-spline fitting to obtain the spline curve. ; The general mathematical expression for a B-spline curve is:

[0037] in, For the i-th control point, Let u be the k-th degree B-spline basis function, and u be the curve parameter. ; Using cubic quasi-uniform B-splines (k=3), basis functions It is given by the following recursive formula:

[0038] in, node vector The value of the i-th node in the vector is uniformly distributed and satisfies the following conditions: , To control the number of points, Given the degree of the B-spline, the recursive starting point for the first-order basis functions is: ; S56. The Douglas-Puk algorithm is used to fit the spline curves. Discretization is performed: a distance threshold is set. Connect the first and last points of the spline curve to form a straight line segment, and calculate the perpendicular distance from all intermediate points on the curve to this straight line segment. If the perpendicular distance from all midpoints to the line segment is less than the distance threshold If the first and last points are not selected, all intermediate points will be discarded, and only the first and last points will be retained. If there exists a point whose perpendicular distance from the midpoint to the line segment is greater than or equal to the distance threshold. Then, retain the point with the largest vertical distance, and use this point as the dividing point to divide the curve into two segments. Recursively perform the above process on each segment until the vertical distance from all intermediate points in each segment to the corresponding line segment is less than the distance threshold. The points that are ultimately retained are the discrete point sequence. ,in The number of discrete points in the m-th layer; In step S56, the specific process of discretization using the Douglas-Puk algorithm is as follows: For a straight line segment formed by connecting the first and last points on a spline curve, calculate the perpendicular distance from all intermediate points on the curve to this straight line segment. If the maximum vertical distance is less than the distance threshold If so, then discard all intermediate points; If the maximum vertical distance is greater than or equal to the distance threshold Then, using this maximum distance point, the curve is divided into two segments. The above process is recursively performed on each of the two segments until the perpendicular distance from all intermediate points to their corresponding line segments is less than the distance threshold. The retained points are the discretized point sequence; S57. For discrete point sequences For each discrete point q in the matrix, its normal vector is calculated using principal component analysis of its local neighborhood point set. ; S58. Sequence of discrete points in each layer and its normal vector Transform to the robot's base coordinate system: Because the industrial robot carries irregularly shaped components and moves relative to the laser beam, the point q in the workpiece coordinate system is transformed into the robot's base coordinate system by the transformation relationship between the local coordinate system established in step S2 and the robot's end flange coordinate system, and then combined with the robot's forward kinematics. This transformation yields the path point in the base coordinate system. ; S59. Determine the spacing between adjacent path points based on the overlap ratio η and the spot diameter D, and determine the robot's moving speed based on the laser pulse frequency f and the point spacing. By combining the normal vector directions of each path point, the robot's end effector posture is determined (ensuring the laser beam direction coincides with the normal vector direction), thus obtaining the complete robot path posture parameters. ; like Figure 6 As shown, the robot path point acquisition algorithm flow is as follows: First, input the feature point cloud of the area to be processed and calculate the average density of the point cloud; determine the slice spacing and slice thickness based on the point cloud density, overlap rate, and spot diameter; construct slice data for each layer based on the strip-shaped bilateral nearest neighbor method; perform uniform downsampling on the slice data; obtain the spline curve by fitting cubic quasi-uniform B-splines; discretize the spline curve using the Douglas-Pock algorithm, set a distance threshold, retain more points at locations with high curvature and fewer points at locations with low curvature; calculate the normal vector of each discrete point; calculate the path points in the robot's base coordinate system by reverse relative calculation; determine the robot's moving speed by combining the overlap rate, spot diameter, and laser pulse frequency; output the complete robot path posture parameters. In step S59, the spacing between adjacent path points is determined based on the overlap rate η and the spot diameter D, i.e., the spacing between adjacent path points = D×(1-η), ensuring that the overlap rate between adjacent spots meets the process requirements. The normal vector direction of each path point is achieved by rotating the robot end posture to make the laser beam axis direction coincide with the normal vector direction. It should be further explained that in step S52, the proportionality coefficient... The value ranges from 0.8 to 1.2, and its value is determined based on the point cloud density and the curvature variation of the component surface: when the surface curvature variation is large, a smaller value is taken to make the slice thickness thinner and capture the surface shape more precisely; when the surface curvature variation is small, a larger value is taken to improve computational efficiency. In step S53, the initial position of the slicing plane (i.e., the starting position of the Z-axis) is set to the plane containing the point with the minimum Z-coordinate in the feature point cloud, and the slicing is advanced sequentially along the positive Z-axis until the entire feature point cloud is covered; the number of slicing layers M is determined by the slicing spacing. The total height H of the feature point cloud along the Z-axis direction determines the following: ; In step S55, the number of control points for the cubic quasi-uniform B-spline curve is determined by the number of slice point clouds and the shape complexity: too few control points will lead to underfitting of the curve and failure to accurately represent the slice shape; too many control points will lead to overfitting of the curve and increase computational redundancy; the preferred number of control points is 1 / 10 to 1 / 5 of the number of slice point clouds, and at least 4. In step S56, the distance threshold ε ranges from 0.02mm to 0.1mm, and its value is determined according to the processing accuracy requirements: when the accuracy requirement is high, a smaller value is used to retain more discrete points to ensure path accuracy; when the accuracy requirement is low, a larger value is used to reduce the number of discrete points to improve processing efficiency. In step S58, the specific implementation of coordinate transformation is as follows: First, the pose matrix of the workpiece coordinate system relative to the robot end flange coordinate system is determined according to the local coordinate system established in step S2. Then, the pose matrix is ​​transformed to the robot base coordinate system according to the robot kinematic model, and finally the position and orientation of the robot end in the base coordinate system are obtained. S6. Control the robot to move along the path according to the robot path points, and cooperate with the laser to perform laser shock strengthening processing until the full coverage processing is completed; The specific steps of step S6 are as follows: S61. Receive the robot path posture parameters Π output in step S5; S62. Convert the path posture parameter Π into robot control instructions and send them to the robot controller. The robot control instructions include the joint angle sequence and the movement speed. S63. The robot moves along the path of the area to be processed according to the control instructions. At the same time, the laser controls the opening, closing and power of the laser beam according to the path parameters. The direction of the laser beam always coincides with the normal vector direction of the current path point and works synchronously with the robot to perform laser shock strengthening processing on the surface of the irregular component. like Figure 7 As shown, the laser surface processing system for irregularly shaped components includes an industrial robot 1, an irregularly shaped component 2, a laser 3, a structured light sensor 4, and a control system. The irregularly shaped component 2 is fixed to the end flange of the industrial robot 1, and the industrial robot 1 carries the irregularly shaped component 2 and moves relative to the fixed laser 3. The structured light sensor 4 is set in the working space of the industrial robot 1 to collect three-dimensional point cloud data of the surface of the irregularly shaped component 2. The laser 3 controls the direction and power of the laser beam according to the path parameters and works synchronously with the industrial robot 1. The control system includes a point cloud acquisition module, a registration module, a depth completion module, a region segmentation module, a path planning module, and a robot control module. Each module is integrated into an industrial control computer and communicates with the controller of the industrial robot 1 through the industrial Ethernet protocol. S64. During the laser shock strengthening process, path tracking monitoring confirms that the robot has traversed all path points, ensuring that all parts of the area to be processed have been uniformly strengthened until full coverage processing is completed; It should be further explained that in step S63, the synchronization control between the laser and the robot is achieved through Ethernet or real-time industrial bus, with a communication cycle of no more than 1ms, ensuring that the laser trigger signal matches the robot's position accuracy; the control method for aligning the laser beam direction with the normal vector direction of the path point is: solving the angles of each joint through the robot's inverse kinematics, so that the robot's end-effector posture satisfies that the laser beam axis direction is consistent with the normal vector direction, with an allowable deviation of no more than ±2°; in step S64, path tracking monitoring is achieved by reading the robot encoder position in real time and comparing it with the desired path point. When the position deviation exceeds the allowable tolerance, the robot control system automatically corrects the position or suspends processing and issues an alarm.

[0039] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0040] like Figure 2 As shown, the following is an embodiment of the automatic planning system for laser surface processing of irregularly shaped components provided in this disclosure. This system and the automatic planning method for laser surface processing of irregularly shaped components in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the automatic planning system for laser surface processing of irregularly shaped components, please refer to the embodiments of the automatic planning method for laser surface processing of irregularly shaped components described above.

[0041] The system includes: The point cloud acquisition module is used to acquire two frames of three-dimensional point cloud data from the surface of the irregular component in two different orientations, and to preprocess each frame of three-dimensional point cloud data separately. The registration module is used to input the preprocessed two frames of 3D point cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; The depth completion module is used to input the registered complete component surface topography point cloud into a pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. The region segmentation module is used to input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. The path planning module is used to determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, to slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, to fit the spline curve based on the sliced ​​point cloud, and to discretize the spline curve to obtain the robot path points. The robot control module is used to control the robot to move along the path according to the robot's path points, and to cooperate with the laser to perform laser shock enhancement processing until the full coverage processing is completed.

[0042] This embodiment achieves automatic planning of laser surface processing paths for complex curved and irregularly shaped components through the interactive collaboration of point cloud acquisition module, registration module, depth completion module, region segmentation module, path planning module, and robot control module. It eliminates the need for CAD models and manual teaching. High-fidelity 3D reconstruction, improved region growth segmentation, and process constraint slicing ensure consistent processing coverage and normal incidence.

[0043] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic planning method for laser surface processing paths of irregularly shaped components, characterized in that, Includes the following steps: S1. Acquire two frames of 3D point cloud data of the surface of the irregular component in two different postures, and preprocess each frame of 3D point cloud data separately; S2. Input the two preprocessed frames of 3D point cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; S3. Input the registered complete component surface topography point cloud into the pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. S4. Input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. S5. Determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, fit the spline curve based on the sliced ​​point cloud, and discretize the spline curve to obtain the robot path points. S6. Control the robot to move along the path according to the robot path points, and cooperate with the laser to perform laser shock strengthening processing until the full coverage processing is completed.

2. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 1, characterized in that, The specific steps of step S1 are as follows: S11. Secure the irregularly shaped component to the end flange of the industrial robot; S12. Position the irregular component in the first posture directly in front of the structured light sensor and collect the first frame of three-dimensional point cloud data on the surface of the irregular component. S13. Position the irregular component in the second posture directly in front of the structured light sensor and collect the second frame of three-dimensional point cloud data on the surface of the irregular component. S14. Perform radius filtering, denoising, and downsampling operations on the first frame of 3D point cloud data and the second frame of 3D point cloud data respectively to obtain the first preprocessed point cloud and the second preprocessed point cloud.

3. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 2, characterized in that, The specific steps of step S2 are as follows: S21. Use the first preprocessed point cloud as the source point cloud and the second preprocessed point cloud as the target point cloud; S22. Source point cloud Each point in the target point cloud The algorithm uses a KD-Tree search to find the nearest point, establishes the correspondence between initial points, and sets the maximum number of iterations, the convergence threshold, and the Cauchy scale parameter. The initial value; S23. Based on the current correspondence, construct the distance error function from the point to the surface, and use the Levenberg-Marquardt algorithm to iteratively optimize the transformation matrix. Minimize the following weighted objective function: in, Let be a rotation matrix. It is a translation vector. Source Point Cloud The i-th point in For target point cloud Zhongyu The corresponding nearest point; for The normal vector at point N; N is the number of points in the source point cloud; The Cauchy weighting function is expressed as follows: in, Let be the point-to-surface distance error of the i-th pair of corresponding points. To control the scale parameters of the weighting intensity; S24. Apply the current optimized transformation matrix. Effect on source cloud Update the source point cloud location; S25. Search for the nearest point correspondence between the updated source point cloud and the target point cloud again, and repeat steps S23 to S24 for iterative optimization; S26. Stop iterating when the change in the transformation matrix is ​​less than the convergence threshold or the maximum number of iterations is reached, and output the final transformation matrix. ; S27. Transform the final matrix Effect on source cloud The source point cloud is transformed to the target point cloud coordinate system to obtain a complete point cloud of the component surface morphology. ; S28. Point cloud of component surface morphology Principal component analysis is used to calculate three principal component directions. A local coordinate system is established, aligned with the Z-axis of the robot's end effector flange, with the first principal component direction as the X-axis, the second principal component direction as the Y-axis, and the third principal component direction as the Z-axis. The complete point cloud with this local coordinate system is then output. .

4. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 3, characterized in that, The specific steps of step S3 are as follows: S31. Use an RGB camera to acquire RGB images of the surface of the irregularly shaped component. ; S32. Transform the complete point cloud with a local coordinate system. With RGB images As input data, it is fed into a pre-trained PENet deep completion model; The PENet deep completion model includes an encoder and a decoder. The encoder processes RGB images... and complete point cloud Feature extraction is performed separately, and the decoder fuses the extracted multimodal features to obtain a complete point cloud. The system performs point-by-point completion of depth information missing due to reflections, occlusions, or sensor limitations, and outputs a completed dense point cloud. ; S33. Dense point cloud Output as a high-fidelity 3D topography reconstruction model.

5. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 1, characterized in that, The specific steps of step S4 are as follows: S41. High-fidelity 3D topography reconstruction model Each point in Take the K nearest neighbors in the local neighborhood to form the neighborhood point set. Calculate the neighborhood point set covariance matrix : in, For neighborhood point set The j-th point in For neighborhood point set The center of mass, ; For covariance matrix Eigenvalue decomposition yields three eigenvalues. Arranged in ascending order as follows The corresponding feature vectors are respectively ,satisfy , It is the index of the eigenvalue. ; Take the smallest eigenvalue Corresponding feature vector As a point normal vector ,Right now ; The point is defined as the proportion of the smallest eigenvalue to the sum of all eigenvalues. Local curvature ,Right now ; Spatial coordinates of the junction Generate a feature vector for each point ; S42. High-fidelity 3D topography reconstruction model In the process, all undivided points are used to form a candidate point set, and the curvature of each point is considered. Sort the values ​​in ascending order, select the undivided point with the smallest curvature value as the initial seed point, and add the initial seed point to the seed list. ; S43. From the seed list Select one point from the list as the current seed point. For the current seed point The KD-Tree search is used to find its K nearest neighbors in space, forming a nearest neighbor set. ; S44. For the neighbor set Each neighboring point in The following three constraints will be evaluated sequentially: Spatial distance constraints: Neighboring points To the current seed point European distance Less than the preset first distance threshold ; in, The calculation method is as follows: Take neighboring points To the current seed point The median of the Euclidean distance; Normal vector angle constraint condition: Neighboring points With the current seed point The angle between the normal vectors Less than the preset angle threshold , The calculation method is as follows: In the formula, Neighboring points The normal vector, The current seed point The normal vector; Curvature constraint conditions: Neighboring points curvature Less than the preset curvature threshold ; If all three conditions are met, then the nearest points will be... Add to current area ; S45. Add the new item to the current area. The point is added to the seed list as a new seed point. and the currently processed seed point From the seed list Remove from; S46. Repeat steps S43 to S44 until the seed list is complete. Empty, current region Growth complete, and the current area As a clustering region Output; S47. High-fidelity 3D topography reconstruction model Remove the already segmented points, return to step S42, and repeat until all points are segmented, resulting in multiple clustered regions. ; S48. Based on the preset minimum clustering threshold and maximum number of points threshold Each cluster region was filtered, and those with fewer than [number of points] were removed. or more Within the specified region, clustered regions that meet the threshold range are retained and merged to form the feature point cloud of the region to be processed. .

6. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 5, characterized in that, In step S44, neighboring points It also needs to satisfy the orthogonal distance constraint to the fitting plane, i.e., the nearest point orthogonal distance to its local fitting plane Less than the preset second distance threshold ; orthogonal distance The calculation method is as follows: in, Neighboring points The centroid of the local neighborhood in the plane, where n is the plane normal vector; Second distance threshold The calculation method is as follows: in, The median of the orthogonal distances between all points in the nearest neighbor set. The median error of the orthogonal distance is calculated as follows: In the formula, Let be the orthogonal distance from the nearest neighbor point to the fitted plane. This indicates that the median is taken, and 1.4826 is the consistency constant for converting the absolute deviation of the median to the standard deviation scale.

7. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 6, characterized in that, The specific steps of step S5 are as follows: S51. Calculate the feature point cloud of the region to be processed. Average point cloud density Specifically: For the feature point cloud of the region to be processed with N points ,use The Euclidean distance between points p and q is expressed by... This represents the distance between point p and its nearest neighbor: ， Then feature point cloud Average point cloud density for: S52. Based on average point cloud density Overlap rate and spot diameter Determine the point cloud slicing parameters, including the slice spacing. With slice thickness ; Slice spacing Determined by the spot diameter and overlap ratio: ； slice thickness Based on average point cloud density Set as ; in This is the proportionality coefficient; S53. Construct slice data based on the strip-shaped bilaterally nearest neighbor method to obtain the slice data point set of all layers. Where M is the number of slice layers, This represents the slice data for the m-th layer; S54. For each layer of sliced ​​data Perform uniform downsampling to reduce the amount of data while maintaining the slice shape, resulting in the downsampled slice point set. ; S55. For each downsampled slice point set Smooth curve fitting is performed using cubic quasi-uniform B-spline fitting to obtain the spline curve. ; S56. The Douglas-Puk algorithm is used to fit the spline curves. Discretization is performed: a distance threshold is set. Connect the first and last points of the spline curve to form a straight line segment, and calculate the perpendicular distance from all intermediate points on the curve to this straight line segment. If the perpendicular distance from all midpoints to the line segment is less than the distance threshold If the first and last points are not selected, all intermediate points will be discarded, and only the first and last points will be retained. If there exists a point whose perpendicular distance from the midpoint to the line segment is greater than or equal to the distance threshold. Then, retain the point with the largest vertical distance, and use this point as the dividing point to divide the curve into two segments. Recursively perform the above process on each segment until the vertical distance from all intermediate points in each segment to the corresponding line segment is less than the distance threshold. The points that are ultimately retained are the discrete point sequence. ,in The number of discrete points in the m-th layer; S57. For discrete point sequences For each discrete point q in the matrix, its normal vector is calculated using principal component analysis of its local neighborhood point set. ; S58. Sequence of discrete points in each layer and its normal vector Transform to the robot's base coordinate system: Because the industrial robot carries irregularly shaped components and moves relative to the laser beam, the point q in the workpiece coordinate system is transformed into the robot's base coordinate system by the transformation relationship between the local coordinate system established in step S2 and the robot's end flange coordinate system, and then combined with the robot's forward kinematics. This transformation yields the path point in the base coordinate system. ; S59. Determine the spacing between adjacent path points based on the overlap ratio η and the spot diameter D, and determine the robot's moving speed based on the laser pulse frequency f and the point spacing. By combining the normal vector directions of each path point, the robot's end-effector posture is determined, thus obtaining the complete robot path posture parameters. .

8. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 7, characterized in that, In step S53, the specific steps are as follows: S531. Define the slicing plane E as the XY plane, and establish a local coordinate system by taking the normal vector of the slicing plane E as the Z-axis; S532. Feature point cloud to be sliced As input, points located in the positive half-space of the slice plane E are assigned to the positive point set. ,set up For the point set The point in the middle, ; Points located in the negative half-space of the slice plane E are classified into the negative point set. ,set up negative point set The point in the middle, ; S533. Point Set and negative point set Construct KD-Tree data structures along the X and Y directions respectively to obtain positive data structures. and negative data structures ; S534. For the point set Each point in In negative data structures The nearest point in the search space So that the two points satisfy: S535. Judgment Point With point Is the Euclidean distance less than the specified search radius r? If not, discard the current point and process the set of points. The next point in; If so, proceed to S536; S536. Connection Point and points Given a line segment L, find the intersection point of line segment L and the slicing plane E. : Set the intersection point Located on line segment L, let the parameter t represents the proportion of the intersection point on line segment L. The coordinates of the intersection point can be obtained from the parametric equation of the spatial line: S537. Correct Point Set Repeat steps S534 to S536 for each point to obtain all intersection points under the current slice plane E, which constitute the slice data point set of this layer. ; S538. Along the Z-axis, according to the slice spacing Translate slice plane E, repeat steps S532 to S537, and traverse the entire feature point cloud. Obtain the slice data point set of all layers. , where M is the number of slice layers.

9. The automatic planning method for laser surface processing path of irregularly shaped components according to claim 1, characterized in that, The specific steps of step S6 are as follows: S61. Receive the robot path posture parameters Π output in step S5; S62. Convert the path posture parameter Π into robot control instructions and send them to the robot controller. The robot control instructions include the joint angle sequence and the movement speed. S63. The robot moves along the path of the area to be processed according to the control instructions. At the same time, the laser controls the opening, closing and power of the laser beam according to the path parameters. The direction of the laser beam always coincides with the normal vector direction of the current path point and works synchronously with the robot to perform laser shock strengthening processing on the surface of the irregular component. S64. During the laser shock peening process, path tracking monitoring confirms that the robot has traversed all path points, ensuring that all parts of the area to be processed have been uniformly strengthened until full coverage processing is completed.

10. An automatic planning system for laser surface processing paths of irregularly shaped components, characterized in that, include: The point cloud acquisition module is used to acquire two frames of three-dimensional point cloud data from the surface of the irregular component in two different orientations, and to preprocess each frame of three-dimensional point cloud data separately. The registration module is used to input the preprocessed two frames of 3D point cloud data into the registration algorithm to perform point cloud registration, obtain the complete surface topography point cloud of the component, and establish a local coordinate system on the component surface that is aligned with the Z-axis of the robot end flange; The depth completion module is used to input the registered complete component surface topography point cloud into a pre-trained 3D depth completion neural network to reconstruct the missing depth information and obtain a high-fidelity 3D topography reconstruction model. The region segmentation module is used to input the high-fidelity 3D topography reconstruction model into the improved region growing segmentation algorithm to extract the feature point cloud of the region to be processed. The improvement includes at least selecting seed points according to curvature sorting, determining the neighborhood belonging by combining the angle between the normal vectors and the curvature dual threshold, and using KD-Tree to accelerate the search. The path planning module is used to determine the point cloud slicing parameters based on the feature point cloud of the area to be processed, to slice the feature point cloud based on the point cloud slicing parameters to obtain the sliced ​​point cloud, to fit the spline curve based on the sliced ​​point cloud, and to discretize the spline curve to obtain the robot path points. The robot control module is used to control the robot to move along the path according to the robot's path points, and to cooperate with the laser to perform laser shock enhancement processing until the full coverage processing is completed.

Images