Three-dimensional trepanning method for curved-surface workpiece processed by laser

Abstract

The invention discloses a three-dimensional nesting method for a curved surface workpiece in laser processing, provides a solution based on robot tail end three-dimensional camera scanning and LSCM mapping, not only solves the problems of manual dependence and efficiency in a traditional method, but also ensures the processing precision of a complex curved surface through an innovative nesting algorithm, and improves the processing precision of the curved surface. The method has remarkable technical advantages and application value.

Description

Technical Field

[0001] The invention relates to the technical field of laser processing, and more specifically to a three-dimensional nesting method for a curved surface workpiece processed by laser. Background Art

[0002] In the laser processing robot scenario, in order to perform laser processing operations such as etching and cleaning on curved workpieces, the robot first needs to obtain the workpiece model in the processing system, use the model to determine the processing area and generate the processing path.

[0003] To this end, the operator needs to attach multiple marking points on the surface of the workpiece, and manually reconstruct the workpiece in three dimensions with the help of a handheld 3D scanner, and finally import the 3D model containing the surface normal vector information into the processing system. In this process, the 3D model of the workpiece needs to be unified with the robot base coordinate system with the help of a calibration system. After unifying the coordinate system, users usually need to use a marker or marking points on the surface of the workpiece to delineate the processing area, which cannot be conveniently planned and cannot preview the processing effect. At the same time, the existing 3D nesting solutions often use projection or unfolding methods, which cannot effectively process complex surfaces, resulting in serious deformation of the processed graphics and difficulty in ensuring accuracy, so there is room for improvement. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a three-dimensional nesting method for laser-processed curved workpieces, which can realize laser processing preview and obtain high-precision processing graphics to adapt to high-precision processing of complex curved surfaces.

[0005] To achieve the above object, the present invention provides the following technical solutions: A three-dimensional nesting method for a curved workpiece processed by laser, comprising the following steps: S1. Connect the 3D depth camera to the robot, and use the robot to move the 3D depth camera to complete the point cloud acquisition of the curved workpiece; S2, preprocessing the point cloud of the curved workpiece; S3, calculating the normal vector of each point in the point cloud and performing Poisson reconstruction to obtain a triangular patch mesh; S4, performing plane parameterization on the triangular face mesh, mapping the triangular face mesh to the plane parameter domain using LSCM to obtain the plane parameter domain vertices; S5, obtaining a parameter plane mapping texture effect diagram and a triangular mesh surface texture mapping effect diagram according to the plane parameter domain vertices; S6. The graphic processing coordinates are obtained according to the vertices of the plane parameter domain, and a processing path is formed. The robot completes the processing action according to the processing path.

[0006] As a further improvement of the present invention, the point cloud collection corresponding to step S1 includes the following steps: S11, control the robot to move to different postures, collect the calibration plate image at each posture, and record the robot terminal posture matrix; S12, calculating the transformation matrix between the camera coordinate system and the robot coordinate system; S13, transforming the point cloud of the camera coordinate system observed by the three-dimensional depth camera into the robot terminal coordinate system through a transformation matrix; S14, controlling the robot to move to each sampling position along a preset path, recording the robot end pose matrix at each position, and collecting point cloud data of the curved workpiece through a three-dimensional depth camera at each position; S15. Calculate the complete coordinate transformation matrix using the transformation matrix and the robot posture information, perform a unified coordinate transformation on all point cloud data, and convert all point clouds into the robot base coordinate system.

[0007] As a further improvement of the present invention, the pre-processing in step S2 includes the following steps: S21, cropping the point cloud; S22. Downsample the point cloud.

[0008] As a further improvement of the present invention, in step S21, the point cloud is clipped into quadrilaterals.

[0009] As a further improvement of the present invention, the normal vector calculation in step S3 includes the following steps: S311, for each point in the point cloud, search for its K1 nearest neighbor points; S312, by calculating the covariance matrix, performing eigenvalue decomposition on it, the covariance matrix is: Among them, n is the number of points in the neighborhood, p i is the coordinate vector (x, y, z) of the i-th point in the neighborhood, and μ is the centroid of the neighborhood point set; S313. The eigenvector corresponding to the minimum eigenvalue is the required normal vector.

[0010] As a further improvement of the present invention, the normal vector is smoothed after being obtained in S3.

[0011] As a further improvement of the present invention, the smoothing process comprises the following steps: S321. For each point in the point cloud, find its K2 nearest neighbor points; S322, calculating the average normal vector of the nearest neighbor point, which is the normal vector of this point after smoothing.

[0012] As a further improvement of the present invention, the plane parameterization in step S4 includes the following steps: S41, extracting a boundary vertex sequence of the mesh model, and forming a closed loop by connecting the boundary vertices in a sequence; S42, let the total number of boundary vertices be n, and associate each boundary vertex with a parameter value on the interval [0,1] through the normalization parameter t = i / (n-1) (i is the vertex number, i∈[0,n-1]); S43. Map the boundary vertices to the four sides of the unit square based on the value of parameter t.

[0013] As a further improvement of the present invention, the mapping rule corresponding to step S43 is: When t∈[0,0.25), the vertex is mapped to the bottom edge of the square with coordinates (0,4t); When t∈[0.25,0.5), the vertex is mapped to the right side of the square with coordinates (4(t-0.25),1); When t∈[0.5,0.75), the vertex is mapped to the upper edge of the square with coordinates (1,1-4(t-0.5)); When t∈[0.75,1], the vertex is mapped to the left side of the square with coordinates (1-4(t-0.75),0).

[0014] Beneficial effects of the present invention: 1. The solution of 3D camera scanning and LSCM mapping at the end of the robot, combined with the nesting algorithm, can achieve high-precision processing of curved surface processing graphics; 2. By connecting the 3D depth camera to the robot, the surface point cloud of the workpiece can be automatically and quickly obtained, eliminating the need to attach marking points to the workpiece surface. The laser processing area can be planned and the processing graphics can be mapped onto the 3D surface, allowing users to preview the laser processing effect in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 This is a point cloud acquisition flow chart of the present invention; Figure 3 A schematic diagram of the robot of the present invention cooperating with a three-dimensional depth camera to obtain a workpiece point cloud; Figure 4 This is a schematic diagram of the point cloud before clipping of the present invention; Figure 5 This is a schematic diagram of the point cloud after clipping in the present invention; Figure 6 This is a Poisson reconstruction grid model diagram of the present invention; Figure 7 It is a schematic diagram of the grid plane parameterization of the present invention; Figure 8 It is the parameter plane mapping texture effect diagram of the present invention; Fig. 9 This is a triangular mesh surface texture mapping effect diagram of the present invention; Fig.10 This is a schematic diagram of the rendering effect based on LSCM graphics nesting according to the present invention. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments, wherein the same components are represented by the same reference numerals.

[0017] like Figure 1-10 As shown, a three-dimensional nesting method for laser-processed curved workpieces is used to perform high-precision processing on the graphic surface of the curved workpiece, including the following steps.

[0018] S1. Connect the 3D depth camera to the robot, and use the robot to move the 3D depth camera to complete the point cloud collection of the curved workpiece.

[0019] The point cloud collection corresponding to step S1 includes the following steps.

[0020] S11, control the robot to move to different postures, collect calibration plate images at each posture, and record the robot terminal posture matrix, specifically collect calibration plate graphics at each posture, preferably 20 pairs.

[0021] S12. Calculate the transformation matrix between the camera coordinate system and the robot coordinate system. The transformation matrix is ​​the hand-eye calibration matrix.

[0022] S13. Convert the point cloud of the camera coordinate system observed by the three-dimensional depth camera to the robot terminal coordinate system through the transformation matrix.

[0023] S14, controlling the robot to move to each sampling position along a preset path, recording the robot end pose matrix at each position, and collecting the surface workpiece point cloud data through a three-dimensional depth camera at each position.

[0024] S15. Calculate the complete coordinate transformation matrix using the transformation matrix and the robot posture information, perform a unified coordinate transformation on all point cloud data, and convert all point clouds into the robot base coordinate system.

[0025] S2. Preprocessing the point cloud of the curved workpiece.

[0026] The preprocessing in step S2 includes the following steps.

[0027] S21, cropping the point cloud, specifically cropping the large-format point cloud into a workpiece area.

[0028] Preferably, the cropped working area is a quadrilateral so as to obtain two pairs of mutually perpendicular boundaries.

[0029] S22. Downsample the point cloud.

[0030] Downsampling can reduce the amount of data and retain valid information, thereby improving the efficiency of later calculations. In this embodiment, voxel grid downsampling is adopted, which specifically includes the following steps: S221. Divide the 3D space into a cubic grid of fixed size.

[0031] S222, traverse all points and calculate which voxel each point belongs to.

[0032] S223. Use a spatial hash table to store voxels and the points contained therein. For each non-empty voxel, calculate the centroid of all points therein, and replace all points in the voxel with the centroid.

[0033] S3. Calculate the normal vector of each point in the point cloud and perform Poisson reconstruction to obtain a triangular patch mesh.

[0034] The calculation of the normal vector includes the following steps.

[0035] S311 . For each point in the point cloud, search for its K1 nearest neighboring points. In this embodiment, K1 is preferably 20.

[0036] S312, by calculating the covariance matrix, performing eigenvalue decomposition on it, the covariance matrix is Among them, n is the number of points in the neighborhood, p i is the coordinate vector (x, y, z) of the i-th point in the neighborhood, and μ is the centroid of the neighborhood point set.

[0037] S313. The eigenvector corresponding to the minimum eigenvalue is the required normal vector.

[0038] Furthermore, after obtaining the normal vector, a smoothing process is performed on it, and the smoothing process specifically includes the following steps.

[0039] S321. For each point in the point cloud, find its K2 nearest neighboring points. In this embodiment, K2 is 10.

[0040] S322, calculating the average normal vector of the nearest neighbor point, which is the normal vector of this point after smoothing.

[0041] Smoothing can reduce the impact of noise and make normal vector changes smoother to improve reconstruction quality.

[0042] S4. Plane parameterization of the triangular patch mesh is performed, and the triangular patch mesh is mapped to the plane parameter domain using LSCM. The plane parameter domain vertices formed correspond to the triangular mesh surface vertices one by one to obtain the graphics processing coordinates in the robot base coordinate system. The LSCM (Least Squares Conformal Maps) method can achieve the mapping of 3D mesh to 2D plane by minimizing the deformation energy. At the same time, there is no need to fix the boundary points, so a more natural unfolding result can be obtained, which effectively reduces the distortion of the edge area.

[0043] The LSCM mapping is as follows.

[0044] The conformal mapping from plane X to surface U requires that the tangent vector X(u,v) must not only maintain orthogonality but also maintain isometry after mapping, satisfying the basic equation of the Cauchy-Riemann equation (1-1).

[0045] (1-1) In order to construct a parameterized linear solution system, complex numbers are used to express the u and v coordinates. For each facet in the triangular mesh surface, a local orthogonal coordinate system is defined, where the local coordinates of the three vertices are recorded as (x1, y1), (x2, y2) and (x3, y3). The process of mapping the surface point U (x, y) to the parameter plane (u, v) can be represented by the discretized Cauchy-Riemann equation (1-2).

[0046] (1-2) Express X and U in complex form, that is, X=x+iy, U=u+iv, formula (1-2) can be converted into formula (1-3).

[0047] (1-3) For each triangle patch, the energy function shown in equation (1-4) is defined, and the least squares conformal parameterization is to find the minimum value of this energy function.

[0048] (1-4) Among them, A T Represents the area of ​​the current triangle T. Then the energy function C(U) in the target mesh area is the sum of the energy functions of all single triangles, as shown in (1-5).

[0049] (1-5) Therefore, the energy function equation of all triangular faces is shown in formula (1-6).

[0050] (1-6) where d T is twice the area of ​​triangle T, U j1 , U j2 , U j3 Represents the three vertices corresponding to the jth triangle, W j1 , W j2 , W j3 is defined as: (1-7) Since the energy function C(U) is the quadratic form of the complex number U1,…,Un, it can also be expressed in matrix form, as shown in formula (1-8).

[0051] (1-8) in is an n×n Hermitian Gram matrix, U* represents the Hermitian conjugate matrix of U, which can be shown as (1-9).

[0052] (1-9) Assume that the number of triangles contained in the triangular mesh surface is m, and the number of vertices is n. In order to find the solution to the optimization problem, two or more points need to be added as fixed points. Assume that p vertices are determined as fixed points, then ,in , so the matrix It is divided into two parts, as shown in formula (1-10).

[0053] (1-10) in, M f is a matrix of size m×(np), M p is an m×p matrix. Therefore, equation (1-8) can be rewritten as shown in equation (1-11).

[0054] (1-11) Finally, the linear system is shown in formula (1-12).

[0055] (1-12) in.

[0056] (1-13) By solving the least squares numerical solution of this equation, the UV coordinates of all vertices of the target triangular mesh surface parameterized to the plane domain based on the LSCM method can be calculated. In order to solve this equation, it is necessary to ensure that the matrix A is full rank, and must have two or more constrained fixed points, that is, p is greater than or equal to 2.

[0057] Since in this embodiment, the cropped point cloud has a pair of parallel boundaries, this embodiment adopts a solution of mapping the boundary points to a square parameter plane.

[0058] The plane parameterization specifically includes the following steps.

[0059] S41, extracting a boundary vertex sequence of the mesh model, and forming a closed loop by connecting the boundary vertices in a sequence.

[0060] S42. Assume that the total number of boundary vertices is n. Then, each boundary vertex can be associated with a parameter value in the interval [0,1] through the normalization parameter t = i / (n-1) (i is the vertex number, i∈[0,n-1]).

[0061] S43. Map the boundary vertices to the four sides of the unit square based on the value of parameter t.

[0062] When t∈[0,0.25), the vertex is mapped to the bottom edge of the square with coordinates (0,4t).

[0063] When t∈[0.25,0.5), the vertex is mapped to the right side of the square with coordinates (4(t-0.25),1).

[0064] When t∈[0.5,0.75), the vertex is mapped to the upper edge of the square with coordinates (1,1-4(t-0.5)).

[0065] When t∈[0.75,1], the vertex is mapped to the left side of the square with coordinates (1-4(t-0.75),0).

[0066] The above mapping scheme evenly distributes the boundaries of the original grid on the target square boundary, ensuring the continuity and uniformity of the boundary. By fixing the positions of these boundary points, a stable boundary constraint is provided for the LSCM parameterization, which is conducive to obtaining a more controllable parameterization result.

[0067] S5. Obtain a parameter plane mapping texture effect map and a triangle mesh surface texture mapping effect map according to the plane parameter domain vertices, thereby obtaining a preview effect according to the plane mapping texture effect map and the triangle mesh surface texture mapping effect map.

[0068] S6. The graphics processing coordinates are obtained according to the vertices of the plane parameter domain, and a processing path is formed. The robot completes the processing action according to the processing path. Through this nesting algorithm, the mapping scheme of LSCM can be used to achieve ultra-high precision processing needs.

[0069] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A three-dimensional nesting method for curved workpieces processed by laser, characterized in that: The following steps are involved: S1. Connect the 3D depth camera to the robot, and use the robot to move the 3D depth camera to complete the point cloud acquisition of the curved workpiece; S2, preprocessing the point cloud of the curved workpiece; S3, calculating the normal vector of each point in the point cloud and performing Poisson reconstruction to obtain a triangular patch mesh; S4, performing plane parameterization on the triangular face mesh, mapping the triangular face mesh to the plane parameter domain using LSCM to obtain the plane parameter domain vertices; S5, obtaining a parameter plane mapping texture effect diagram and a triangular mesh surface texture mapping effect diagram according to the plane parameter domain vertices; S6. The graphic processing coordinates are obtained according to the vertices of the plane parameter domain, and a processing path is formed. The robot completes the processing action according to the processing path.

2. A three-dimensional nesting method for laser-processed curved workpieces according to claim 1, characterized in that: The point cloud collection corresponding to step S1 includes the following steps: S11, control the robot to move to different postures, collect the calibration plate image at each posture, and record the robot terminal posture matrix; S12, calculating the transformation matrix between the camera coordinate system and the robot coordinate system; S13, transforming the point cloud of the camera coordinate system observed by the three-dimensional depth camera into the robot terminal coordinate system through a transformation matrix; S14, controlling the robot to move to each sampling position along a preset path, recording the robot end pose matrix at each position, and collecting point cloud data of the curved workpiece through a three-dimensional depth camera at each position; S15. Calculate the complete coordinate transformation matrix using the transformation matrix and the robot posture information, perform a unified coordinate transformation on all point cloud data, and convert all point clouds into the robot base coordinate system.

3. A three-dimensional nesting method for laser-processed curved workpieces according to claim 1, characterized in that: The preprocessing in step S2 includes the following steps: S21, cropping the point cloud; S22. Downsample the point cloud.

4. A three-dimensional nesting method for laser-processed curved workpieces according to claim 3, characterized in that: In step S21, the point cloud is clipped into quadrilaterals.

5. A three-dimensional nesting method for laser-processed curved workpieces according to claim 1, characterized in that: The normal vector calculation in step S3 includes the following steps: S311, for each point in the point cloud, search for its K1 nearest neighbor points; S312, by calculating the covariance matrix, performing eigenvalue decomposition on it, the covariance matrix is: Among them, n is the number of points in the neighborhood, p i is the coordinate vector (x, y, z) of the i-th point in the neighborhood, and μ is the centroid of the neighborhood point set; S313. The eigenvector corresponding to the minimum eigenvalue is the required normal vector.

6. A three-dimensional nesting method for laser-processed curved workpieces according to claim 1, characterized in that: After the normal vector is obtained in S3, it is smoothed.

7. A three-dimensional nesting method for laser-processed curved workpieces according to claim 6, characterized in that: The smoothing process consists of the following steps: S321, for each point in the point cloud, find its K2 nearest neighbor points; S322, calculating the average normal vector of the nearest neighbor point, which is the normal vector of this point after smoothing.

8. A three-dimensional nesting method for laser-processed curved workpieces according to claim 4, characterized in that: The plane parameterization in step S4 includes the following steps: S41, extracting a boundary vertex sequence of the mesh model, and forming a closed loop by connecting the boundary vertices in a sequence; S42, let the total number of boundary vertices be n, and associate each boundary vertex with a parameter value on the interval [0,1] through the normalization parameter t = i / (n-1) (i is the vertex number, i∈[0,n-1]); S43. Map the boundary vertices to the four sides of the unit square based on the value of parameter t.

9. A three-dimensional nesting method for laser-processed curved workpieces according to claim 8, characterized in that: The mapping rule corresponding to step S43 is: When t∈[0,0.25), the vertex is mapped to the bottom edge of the square with coordinates (0,4t); When t∈[0.25,0.5), the vertex is mapped to the right side of the square with coordinates (4(t-0.25),1); When t∈[0.5,0.75), the vertex is mapped to the upper edge of the square with coordinates (1,1-4(t-0.5)); When t∈[0.75,1], the vertex is mapped to the left side of the square with coordinates (1-4(t-0.75),0).

Images