Robot deburring machining path correction method and device based on non-rigid registration

Abstract

The present application belongs to the technical field of robot automation deburring, and discloses a robot deburring processing path correction method and equipment based on non-rigid registration, which comprises the following steps: (1) calculating the local covariance matrix of points in the point cloud data of an actual workpiece, and then dividing the corresponding points into burr points or workpiece surface points; (2) performing Poisson surface reconstruction on the workpiece surface point set to expand the boundary points, and constructing a first search space; performing adjacent point search on the points in the burr point set in the first search space to obtain burr height; (3) taking the point cloud data of the actual workpiece as a target point set, deforming the CAD model of the workpiece by using non-rigid registration, and calculating the transformation relationship between the CAD model and the actual workpiece; (4) generating a theoretical processing path by using the CAD model, adjusting the theoretical processing path by using the transformation relationship, and generating a final processing path of the actual workpiece. The present application improves the deburring path precision.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of robot automatic deburring, and more particularly relates to a robot deburring processing path correction method and device based on non-rigid registration. BACKGROUND

[0002] Most industrial cast parts have burrs and need to be removed. The tolerance of the cast parts is large, the size is not uniform, and the profile position of the burr end surface of the blank exists position error and shape deviation compared with the theoretical model. There is a large error between the trajectory in the robot deburring process and the actual model. Therefore, the robot motion trajectory needs to be determined according to the actual part feature size position.

[0003] At present, the automatic robot deburring processing path generation adjustment method mainly includes: adjusting the tool path based on teaching point matching, extracting the processing path based on two-dimensional vision, and adjusting the path based on three-dimensional point cloud rigid registration. Among them, the method of adjusting the path based on teaching point matching has limited number of teaching points, and the adjustment error is affected by the number of points, and a lot of manual time is consumed. Since the part profile shape and tolerance deformation direction are random, the method of extracting the processing path based on two-dimensional vision is only suitable for simple planar workpieces with simple shapes, and cannot be applied to three-dimensional workpiece machining with complex shapes. The three-dimensional point cloud registration method basically adopts the rigid registration method to adjust the path. However, uneven cold shrinkage and deformation are easily generated in the casting process, causing large tolerance between the cast part and the design model, and large individual differences of the actual part, which leads to inaccurate matching relationship between the actual workpiece point cloud and the design model, and the rigid registration cannot adapt to the deformation of the part individual. Therefore, a new robot deburring processing path correction method needs to be designed.

[0004] Through the above analysis, the problems and defects of the prior art are:

[0005] (1) The existing tool path adjustment method based on teaching point matching has randomness and large error, and consumes a lot of manual time.

[0006] (2) The existing two-dimensional vision-based processing path extraction method cannot be applied to three-dimensional workpiece machining with complex shapes.

[0007] (3) The existing three-dimensional point cloud rigid registration adjustment path method depends on the corresponding relationship between the actual workpiece and the design model, and is difficult to be applied to cast parts with large individual tolerance differences. SUMMARY

[0008] In view of the above defects or improvement needs of the prior art, the present application provides a robot deburring machining path correction method and device based on non-rigid registration, which corrects the deburring machining path by using a non-rigid registration method considering the burr height, reduces the deburring machining errors caused by part tolerances and burrs, and improves the deburring path precision.

[0009] To achieve the above-mentioned purpose, according to one aspect of the present application, a robot deburring machining path correction method based on non-rigid registration is provided, which comprises the following steps:

[0010] (1) calculating the local covariance matrix of the points in the point cloud data of the actual workpiece for representing the bending degree of the curved surface formed by the point cloud near the corresponding point; then, calculating the eigenvalues of the local covariance matrix, and then calculating the feature quantity from the eigenvalues, and dividing the corresponding point into a burr point or a workpiece surface point according to the ratio of the feature quantity to the set threshold value, and then obtaining the burr point point set and the workpiece surface point point set;

[0011] (2) performing Poisson surface reconstruction on the workpiece surface point point set to expand the boundary points, and constructing a first search space based on the reconstructed surface point point set; then, performing adjacent point search on the points in the burr point point set in the first search space to obtain the nearest point distance, and the nearest point distance corresponding to the burr point being the burr height thereof;

[0012] (3) taking the point cloud data of the actual workpiece as a target point set, deforming the CAD model of the workpiece by using non-rigid registration, and calculating the transformation relationship between the CAD model and the actual workpiece;

[0013] (4) generating a theoretical machining path by using the CAD model, adjusting the theoretical machining path by using the transformation relationship, and generating the final machining path of the actual workpiece.

[0014] Further, before step (1), the method further comprises the step of: using a voxel grid sampling algorithm to down-sample the scanned point cloud data of the actual workpiece, and performing outlier filtering to obtain the point cloud data of the actual workpiece.

[0015] Further, the scanned point cloud data P0 of the original workpiece is down-sampled by using a voxel grid sampling algorithm, a cube that just wraps the point cloud data P0 is found, the cube is divided into a plurality of small cube grids according to a given grid edge length, the centroids of the points in each grid are calculated, and these points are replaced by the centroid points to obtain the down-sampled point cloud data P down ; the down-sampled data P down is taken as a search space, the average distance of k nearest neighbors of each point in P down is searched, if the average distance is greater than a set threshold value, the point is removed, and the outlier filtered point cloud data P1 is obtained.

[0016] Further, for the point cloud data P1 of the actual workpiece, a search space is constructed, and points p in the point cloud data P1 are searched in the space i k nearest neighbors of the point p The k points searched constitute a set

[0017]

[0018] Wherein, p i is a point in the point cloud data P1, is the centroid of the point set is the local covariance matrix at the point p i ;

[0019] The eigenvalues of the matrix are calculated, since is symmetric and positive semi-definite, the three eigenvalues of are positive real numbers, λ1≥λ2≥λ3≥0;

[0020] The characteristic quantity is defined to represent the bending degree of the curve fitted by the point set in the neighborhood of the point p i , which is equivalent to the surface change rate at the point p i ;

[0021] When σ k (p i ) exceeds a set threshold, the point p i is a boundary burr point, otherwise it is a workpiece surface point; for each point in the point cloud data P1, calculation is performed to obtain the boundary burr point set P inliers and the workpiece surface point set P outliers .

[0022] Further, take the template S = {V, E} of the CAD nominal model, which contains the source point set V and the edge set E, for each point v on the CAD model i , find the corresponding point p i on the actual workpiece point cloud scanning model using nearest neighbor search, and define the weight term coefficient and calculate according to the burr height d burr of the point p i .

[0023] For non-burr points, w i = 1.

[0024] Further, according to the calculation result of the weight coefficient, the target function of registration is calculated: ​

[0025]

[0026]

[0027] E(T)=E d (T)+αE s (T)

[0028] Wherein T i is the transformation matrix of point v i , T=[T1,T2…T n ] T is the parameter to be solved, dist(p1,p2) is the Euclidean distance between points p1 and p2;G=diag(1,1,1,k s ), k s is used to weight the difference between the rotational distortion part and the translation part of the deformation;Alpha is the stiffness weight;E(T) is the target cost function to be minimized.

[0029] Further, by setting the derivative of E(T) to zero, the linear equation system obtained by solving obtains T when E(T) is minimum;

[0030] Repeat the solution of E(T), and reduce the value of alpha, until the maximum iteration number is reached or the error is less than the set value, to obtain the final transformation relationship T.

[0031] Further, the CAD model is generated to generate a deburring processing path, path points are generated at the edge contour of the part prone to burr, and the transformation relationship obtained by non-rigid registration is calculated to correct the tool position point:

[0032]

[0033] Wherein, the transformation matrix at the tool position point T i is T iTrans , and the corrected tool position point is calculated Convert each tool position path point and posture to obtain the final processing path.

[0034] The application also provides a robot deburring processing path correction system based on non-rigid registration, the system comprises a data preprocessing module, a boundary burr point set segmentation module, a burr height calculation module, a non-rigid registration calculation module and a target path correction adjustment module;Wherein,

[0035] The data preprocessing module is used for reducing the sampling of the actual workpiece point cloud data by using the voxel grid sampling algorithm, and performing outlier filtering to obtain the point cloud data of the actual workpiece;

[0036] The boundary burr point set segmentation module is used to calculate the local covariance matrix of points in the point cloud data of the actual workpiece, so as to characterize the bending degree of the curved surface formed by the point cloud near the corresponding point; then, the eigenvalues of the local covariance matrix are calculated, and then the feature quantity is calculated from the eigenvalues, and the corresponding point is divided into a burr point or a workpiece surface point according to the ratio of the feature quantity to the set threshold, so as to obtain the burr point set and the workpiece surface point set;

[0037] The burr height calculation module is used to perform Poisson surface reconstruction on the workpiece surface point set to expand the boundary points, and construct a first search space based on the reconstructed surface point set; then, the points in the burr point set are searched for adjacent points in the first search space to obtain the nearest point distance, and the nearest point distance corresponding to the burr point is the burr height of the burr point;

[0038] The non-rigid registration calculation module is used to take the point cloud data of the actual workpiece as a target point set, deform the CAD model of the workpiece by using non-rigid registration, and calculate the transformation relationship between the CAD model and the actual workpiece;

[0039] The target path correction and adjustment module is used to generate a theoretical machining path by using the CAD model, adjust the theoretical machining path by using the transformation relationship, and generate a final machining path of the actual workpiece.

[0040] The application also provides a computer storage medium, which stores machine executable instructions, and the machine executable instructions make the processor realize the non-rigid registration based robot deburring machining path correction method as described above when the processor calls and executes the machine executable instructions.

[0041] Overall, compared with the prior art, the non-rigid registration based robot deburring machining path correction method and device provided by the application mainly has the following beneficial effects:

[0042] 1. The application takes into account the local deformation of the cast workpiece, accurately calculates the transformation relationship between the design CAD model and the actual part by using non-rigid registration, and corrects the tool path by using the transformation relationship, so that the correction of the theoretical tool path is realized in the case that the local deformation of the workpiece is unknown.

[0043] 2. By measuring the burr height, the burr height is added as a registration weight item, the influence of the burr on the registration itself is avoided, the under-cut phenomenon that may be caused by the influence of the final path on the burr noise is prevented, and the accuracy of the final corrected path is improved.

[0044] 3. The application uses a non-rigid registration method, adds a registration method considering the influence of burrs, iteratively calculates the final change matrix, reflects the deformation of the workpiece relative to the design CAD model, and reduces the deburring machining error caused by the existence of part tolerance and burr compared with the traditional rigid registration scheme, and improves the accuracy of the deburring path. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a flowchart of the robot deburring machining path correction method based on non-rigid registration provided by the application;

[0046] Figure 2 is a principle diagram of the robot deburring machining path correction method based on non-rigid registration provided by the application;

[0047] Figure 3 is a structure block diagram of the robot deburring machining path correction system based on non-rigid registration provided by the application;

[0048] Figure 4 (a), (b), (c) in are respectively the profile error comparison diagrams of the machining path and the whole rigid registration and local rigid registration corrected machining path provided by the embodiment of the application.

[0049] In all the drawings, the same reference signs are used to represent the same elements or structures, wherein: 1-data preprocessing module, 2-boundary burr point set segmentation module, 3-burr height calculation module, 4-non-rigid registration calculation module, 5-target path correction adjustment module. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages of the application more clear and obvious, the application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application. In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as there is no conflict.

[0051] Example 1

[0052] Please refer to Figure 1 and Figure 2 In order to solve the problem that the existing method is difficult to consider the local tolerance deformation of the workpiece in removing burrs, the application provides a robot deburring machining path correction method based on non-rigid registration, which aims to reduce the deburring path error caused by tolerance deformation and burrs and improve the deburring machining accuracy.

[0053] The method mainly includes the following steps:

[0054] S101, data preprocessing is performed on the measurement point cloud, noise is removed, and the data quantity is reasonable and more uniform.

[0055] The original workpiece scanning point cloud data is down-sampled by using a voxel grid sampling algorithm, and outlier filtering is performed to obtain the point cloud data of the actual workpiece.

[0056] Specifically, the original workpiece scanning point cloud data P0 (measurement point cloud data) is down-sampled by using a voxel grid sampling algorithm, a cube that just wraps the point cloud data P0 is found, the cube is divided into a plurality of small cube grids according to a given grid side length, the centroid of the points in each grid is calculated, and these points are replaced by the centroid points to obtain the down-sampled point cloud data P down The down-sampled data P down is taken as a search space, and the average distance of k nearest neighbors of each point in P down is searched, if the average distance is greater than a set threshold, the point is removed, and the outlier filtered point cloud data P1 is obtained.

[0057] S102, the boundary burr point set and the other point set are segmented according to the point cloud features.

[0058] The local covariance matrix of the points in the point cloud data of the actual workpiece is calculated to represent the bending degree of the surface formed by the point cloud near the corresponding point; then, the eigenvalues of the local covariance matrix are calculated, and then the feature quantity is calculated from the eigenvalues, and the corresponding point is divided into a burr point or a workpiece surface point according to the ratio of the feature quantity to the set threshold, and then the burr point set and the workpiece surface point set (i.e. other point set) are obtained.

[0059] Specifically, for the point cloud data P1 of the actual workpiece, a search space is constructed and stored in a kdTree structure. The k nearest neighbors of a point p i in the point cloud data P1 are searched in the space, and the k points searched constitute a set

[0060] The local covariance matrix of the points in the point cloud data P1 is calculated to represent the bending degree of the surface formed by the point cloud near the corresponding point:

[0061]

[0062] Where p i is a point in the point cloud data P1, is the centroid of the point set , and is the local covariance matrix of the point p i , and is a symmetric positive semi-definite matrix.

[0063] The matrix The eigenvalues of the characteristic equation, since Symmetric semi-positive definite, it is known that The three eigenvalues of the characteristic equation are positive real numbers. Assume that the eigenvalue obtained is λ 1，2，3 , where λ1≥λ2≥λ3≥0.

[0064] The characteristic quantity is defined as The point p i The bending degree of the point set fitting surface in the neighborhood of the point p i is equivalent to the surface change rate at the point p k .

[0065] According to the size of the characteristic quantity σ i (p k ), a threshold value is set. When σ i (p i ) exceeds the threshold value, the point p inliers is a boundary burr point, and otherwise it is a workpiece surface point. The calculation is performed for each point in the point cloud data P1, and the boundary burr point set P outliers and the workpiece surface point set P outliers are obtained.

[0066] S103, the missing part of the boundary point is reconstructed, the boundary point set and the burr point set are segmented, and the burr height is calculated.

[0067] The workpiece surface point set is Poisson surface reconstructed to expand the boundary point, and a first search space is constructed based on the reconstructed surface point set; then, the points in the burr point set are searched for neighboring points in the first search space to obtain the nearest point distance, and then the corresponding points are divided into burr points or boundary points based on the ratio of the nearest point distance to the threshold value; wherein the nearest point distance corresponding to the burr point is the burr height of the burr point.

[0068] Specifically, the workpiece surface point set P outliers is Poisson surface reconstructed to expand the boundary point. The reconstructed point set is constructed into a first search space, and is stored in a kdTree structure.

[0069] For the points in the boundary burr point set P inliers , the nearest neighbor point search is performed in the first search space to obtain the nearest point distance d burr . If the nearest point distance d burr is greater than a set threshold value M d , the point is a burr point, and d burr is the burr height of the point, and otherwise it is a boundary point.

[0070] S104, the actual point cloud data is taken as a target point set, and a non-rigid registration method is used to deform the CAD model.

[0071] The point cloud data of the actual workpiece is used as the target point set. The CAD model of the workpiece is deformed using non-rigid registration, and the transformation relationship between the CAD model and the actual workpiece is calculated.

[0072] Specifically, we take the template S = {V, E} of the CAD nominal model, where V contains the set of source points and E contains the set of edges. For each point v on the CAD model... i The corresponding point p is found on the actual workpiece point cloud scanning model using nearest neighbor search. i According to p i The height d of the burr at the point burr Define and calculate the coefficients of the weighting terms.

[0073]

[0074] For non-burr points, w i =1.

[0075] Based on the calculation results of the weighting coefficients, the objective function for registration is calculated.

[0076]

[0077]

[0078] E(T)=E d (T)+αE s (T)

[0079] Where T i For point v i The transformation matrix, T = [T1, T2…T n ] T Here are the parameters to be determined, and dist(p1,p2) is the Euclidean distance between points p1 and p2; G = diag(1,1,1,k) s ), k s It can be used to weight the difference between the rotational and translational components of the deformation. α is the stiffness weight, which affects the flexibility of the template deformation. E(T) is the objective cost function that needs to be minimized.

[0080] By setting the derivative of E(T) to zero, the linear equation system obtained by solving the equations can accurately yield T when E(T) is minimized.

[0081] Repeat the above steps and decrease the value of α until the maximum number of iterations is reached or the error is less than the set value, to obtain the final transformation relationship T.

[0082] S105 uses the CAD model to generate the theoretical machining path, adjusts the theoretical machining path using transformation relationships, and generates the final machining path for the actual workpiece.

[0083] The CAD model generation theory is used to generate the deburring machining path, and path points are generated at the edge contour of the part prone to burrs. According to the transformation relationship obtained by non-rigid registration, the corrected tool position point is calculated:

[0084]

[0085] The transformation matrix at the tool position point T i is T iTrans , and the corrected tool position point is calculated as The path point and posture of each tool position point are converted to obtain the final machining path.

[0086] The present application adds a registration coefficient considering the influence of burrs by using a non-rigid registration method, iteratively calculates the final transformation matrix, and reflects the deformation of the actual workpiece relative to the designed CAD model. Compared with the traditional rigid registration scheme, the deburring machining error caused by the part tolerance and burr is reduced, and the deburring path precision is improved.

[0087] Referring to Figure 3 , the present application also provides a robot deburring machining path correction system based on non-rigid registration, which comprises a data preprocessing module 1, a boundary burr point set segmentation module 2, a burr height calculation module 3, a non-rigid registration calculation module 4 and a target path correction and adjustment module 5. Among them,

[0088] The data preprocessing module 1 is used to reduce the sampling of the actual workpiece's point cloud data by using the voxel grid sampling algorithm, and to filter the outliers to obtain the point cloud data of the actual workpiece.

[0089] The boundary burr point set segmentation module 2 is used to calculate the local covariance matrix of the points in the point cloud data of the actual workpiece, so as to represent the bending degree of the surface formed by the point cloud near the corresponding point. Then, the eigenvalues of the local covariance matrix are calculated, and then the feature quantities are calculated from the eigenvalues, and the corresponding points are divided into burr points or workpiece surface points according to the ratio of the feature quantities to the set threshold, and then the burr point set and the workpiece surface point set are obtained.

[0090] The burr height calculation module 3 is used to perform Poisson surface reconstruction on the workpiece surface point set to expand the boundary points, and construct a first search space based on the reconstructed surface point set. Then, the points in the burr point set are searched for adjacent points in the first search space to obtain the nearest point distance, and the nearest point distance corresponding to the burr point is its burr height.

[0091] The non-rigid registration calculation module 4 is used to use the non-rigid registration to deform the CAD model of the workpiece by taking the point cloud data of the actual workpiece as the target point set, and to calculate the transformation relationship between the CAD model and the actual workpiece.

[0092] The target path correction adjustment module 5 is configured to generate a theoretical machining path by using the CAD model, adjust the theoretical machining path by using the transformation relationship, and generate a final machining path of the actual workpiece.

[0093] The application further provides a computer readable storage medium, characterized by: the computer readable storage medium stores machine executable instructions, when the machine executable instructions are called and executed by a processor, the machine executable instructions cause the processor to implement the robot deburring path correction method based on non-rigid registration.

[0094] Embodiment 2

[0095] The application can be used in the automatic machining path generation technology of industrial cast parts, most of which have burrs and need to be removed. The tolerance of the cast parts is large, the size is not uniform, the profile position of the burr end surface of the blank exists position error and shape deviation compared with the theoretical model, and there is a large error between the trajectory in the robot deburring process and the actual model. The application can determine the robot motion trajectory according to the actual part feature size, and improve the deburring machining precision.

[0096] The deburring machining equipment used in the embodiment includes a six-degree-of-freedom robot, a pneumatic spindle, a deburring tool and a machining workbench. The machining workbench is used to carry the machining workpiece, and the pneumatic spindle is connected to the end of the six-degree-of-freedom robot and the deburring tool.

[0097] The data of the measurement point cloud is preprocessed to eliminate the influence of noise, so that the data amount is reasonable and more uniform. The CAD model point cloud is sampled. According to the characteristics of the point cloud, the boundary burr point set and the other point set are segmented from the preprocessed actual model point cloud. Then, the boundary point missing part is reconstructed, the boundary point set and the burr point set are segmented, and the burr height is calculated. The actual point cloud data is taken as a target point set, and the CAD model is deformed by using a non-rigid registration method. The theoretical machining path is generated by using the CAD model, the theoretical machining path is adjusted according to the registration result, and the final machining path suitable for the actual part is generated.

[0098] The machining path is generated by using the method proposed in the application and the whole rigid registration and local rigid registration path correction methods. The theoretical profile of the actual model is compared with the corrected path, and the profile error is calculated. The profile error is compared as shown in (a), (b) and (c) in Figure 4 Figure 4 It can be seen that, compared with other rigid registration methods, the machining path generated by the proposed method effectively reduces the profile error relative to the actual workpiece.

[0099] ​The machining paths generated by the method and the overall rigid registration and local rigid registration correction path method are used to machine the actual parts, measure the chamfer after machining, and calculate the root mean square error of the chamfer on each path. Table 1 below shows the comparison of the root mean square error, and it can be seen that the root mean square error of the chamfer machined by the method is effectively reduced, and the effect of the machined chamfer is more stable and smooth.

[0100] Table 1: Statistics of chamfer root mean square error of parts

[0101]

[0102] In the above embodiments, all or part can be realized by software, hardware, firmware or any combination thereof. When all or part is realized in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL) or wireless (such as infrared, wireless, microwave, etc.)) mode. The computer readable storage medium can be any available medium that the computer can access or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as DVD), or semiconductor media (such as solid state disk Solid State Disk (SSD)) and the like.

[0103] Those skilled in the art will readily understand that the above description is only preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of robot deburring process path correction based on non-rigid registration, characterized in that, The method comprises the following steps: (1) calculating the local covariance matrix of a point in the point cloud data of the actual workpiece, so as to represent the bending degree of the surface formed by the point cloud near the corresponding point; then, the eigenvalue of the local covariance matrix is calculated, and then the characteristic quantity is calculated from the eigenvalue, and the corresponding point is divided into a burr point or a workpiece surface point according to the ratio of the characteristic quantity to the set threshold value, so as to obtain the burr point set and the workpiece surface point set; (2) Poisson surface reconstruction is performed on the workpiece surface point set to expand the boundary points, and a first search space is constructed based on the reconstructed surface point set; then, the points in the burr point set are searched for adjacent points in the first search space to obtain the nearest point distance, and the nearest point distance corresponding to the burr point is the burr height of the burr point; (3) using the point cloud data of the actual workpiece as a target point set, deforming the CAD model of the workpiece by using non-rigid registration, and calculating the transformation relationship between the CAD model and the actual workpiece; (4) generating a theoretical machining path using the CAD model, adjusting the theoretical machining path using the transformation relationship, and generating a final machining path for the actual workpiece.

2. The non-rigid registration based robotic deburring process path correction method of claim 1, wherein: Before step (1), the step of using a voxel grid sampling algorithm to downsample the scanned point cloud data of the actual workpiece, and performing outlier filtering to obtain the point cloud data of the actual workpiece is further included.

3. The non-rigid registration based robotic deburring machining path correction method of claim 2, wherein: The original workpiece scanning point cloud data P0 is sampled by a voxel grid sampling algorithm, is down-sampled, a cube just covering the point cloud data P0 is found, the cube is divided into a plurality of small cube grids according to a given grid side length, the centroid of points in each grid is calculated, the points are replaced by the centroid points, and down-sampled point cloud data P is obtained down ; the down-sampled data P down is taken as a search space, the average distance of k nearest neighbors of each point in P down is searched, if the average distance is greater than a set threshold, the point is removed, and the outlying point filtered point cloud data P1 is obtained.

4. The non-rigid registration based robotic deburring process path correction method of claim 3, wherein: For the point cloud data P1 of the actual workpiece, construct a search space and search for point p in the point cloud data P1 within the space. i The k nearest neighbors are found, and the k nearest neighbors form a set. Calculate the local covariance matrix of a point in the point cloud data P1 to characterize the curvature of the surface formed by the point cloud near that point: wherein p i is a point in the point cloud data P1, is a centroid of the point set , C pi is a local covariance matrix at the point p i . Eigenvalues of the matrix Since Symmetric semi-positive definite, The three eigenvalues of are positive real numbers, λ1≥ λ2≥ λ3≥ 0; Definition of characteristic quantity A curvature of a surface at point p i A curvature of a surface at point p i A curvature of a surface at point p When σ k (p i ) exceeds a set threshold, the point p i is a boundary burr point, otherwise it is a workpiece surface point; the calculation is performed on each point in the point cloud data P1 to obtain a boundary burr point set P inliers and a workpiece surface point set P outliers .

5. The non-rigid registration based robotic deburring process path correction method of claim 4, wherein: Take the template S = {V, E} of the CAD nominal model, where V contains the set of source points and E contains the set of edges. For each point v on the CAD model... i The corresponding point p is found on the actual workpiece point cloud scanning model using nearest neighbor search. i According to p i The height d of the burr at the point burr Define and calculate the coefficients of the weighting terms: For non-spur points, w i = 1.

6. The non-rigid registration based robotic deburring process path correction method of claim 5, wherein: According to the calculation result of the weight coefficient, the target function of registration is calculated: E(T) = E d (T) + aE s (T) where T i is the transformation matrix of point v i , T = [T1, T2, …, T n ] T is the parameter to be solved, dist(p1, p2) is the Euclidean distance between points p1 and p2; G = diag(1, 1, 1, k s ), k s is used to weight the difference between the rotational and translational parts of the deformation; α is the stiffness weight; and E(T) is the target cost function to be minimized.

7. The non-rigid registration based robotic deburring machining path correction method of claim 6, wherein: By setting the derivative of E(T) to zero, the linear equation system obtained by solving E(T) at the minimum E(T) is obtained; Repeat the solution of E(T), and reduce the value of α until the maximum iteration number is reached or the error is less than the set value, to obtain the final transformation relationship T.

8. The non-rigid registration based robotic deburring process path correction method of claim 7, wherein: The theoretical deburring machining path is generated using the CAD model, the path points are generated at the same arc length on the edge contour of the part prone to burrs, and the correction tool position points are calculated according to the transformation relationship obtained by non-rigid registration: T i r = T iTrans x T i Wherein, the transformation matrix of the tool position point T i is T iTrans , the corrected tool position is T i r ; each tool position path point and posture is converted to obtain the final machining path.

9. A non-rigid registration based robotic deburring machining path correction system, characterized by: The system comprises a data preprocessing module, a boundary burr point set segmentation module, a burr height calculation module, a non-rigid registration calculation module, and a target path correction and adjustment module; wherein, The data preprocessing module is used to downsample the scanned point cloud data of the actual workpiece using a voxel grid sampling algorithm, and perform outlier filtering to obtain the point cloud data of the actual workpiece; The boundary burr point set segmentation module is used to calculate the local covariance matrix of a point in the point cloud data of the actual workpiece, so as to represent the bending degree of the surface formed by the point cloud near the corresponding point; then, the eigenvalue of the local covariance matrix is calculated, and then the characteristic quantity is calculated from the eigenvalue, and the corresponding point is divided into a burr point or a workpiece surface point according to the ratio of the characteristic quantity to the set threshold value, so as to obtain the burr point set and the workpiece surface point set; The burr height calculation module is used to perform Poisson surface reconstruction on the workpiece surface point set to expand the boundary points, and construct a first search space based on the reconstructed surface point set; then, the points in the burr point set are searched for adjacent points in the first search space to obtain the nearest point distance, and the nearest point distance corresponding to the burr point is the burr height of the burr point; The non-rigid registration calculation module is configured to use the point cloud data of the actual workpiece as a target point set, deform the CAD model of the workpiece by using non-rigid registration, and calculate a transformation relationship between the CAD model and the actual workpiece; The target path correction and adjustment module is configured to generate a theoretical machining path by using the CAD model, adjust the theoretical machining path by using the transformation relationship, and generate a final machining path of the actual workpiece.

10. A computer storage medium, characterized in that: The computer readable storage medium stores machine executable instructions, and when the machine executable instructions are called and executed by the processor, the machine executable instructions cause the processor to implement the non-rigid registration based robot deburring machining path correction method in any one of claims 1-8.

Images