Grinding method and system of intelligent grinding robot

Abstract

The invention discloses a grinding method and system of an intelligent grinding robot. The method belongs to the field of point cloud processing and machine learning, and comprises the following operation steps: acquiring three-dimensional point cloud data of a polishing target, and acquiring geometric information of the surface of the target through high-precision three-dimensional scanning equipment; preprocessing the point cloud data; polishing target point cloud features are extracted, and the normal direction, the curvature, the height difference and the point cloud density are calculated and combined into feature vectors; machine learning identifies a key geometric region according to the feature vector, and region segmentation is performed on the surface of the polishing target; self-adaptive polishing paths are generated for different areas, optimal polishing tracks are designed for different areas, and motion parameters of the mechanical arm are planned; and automatic polishing operation is executed, and the mechanical arm and polishing equipment are controlled to complete high-precision surface treatment according to the path. The system comprises a three-dimensional information acquisition device, a mechanical arm, a polishing device and a control computer. According to the method, the adaptability to the polishing target is improved by fusing the point cloud processing technology and machine learning.

Description

Technical Field

[0001] The present invention belongs to the field of point cloud processing and machine learning, and relates to a polishing method and system of an intelligent polishing robot. Background Art

[0002] In the existing technology, polishing is one of the steps in the manufacturing industry, and the quality of the target surface directly affects the accuracy and service life of the product. However, the traditional polishing process is highly dependent on manual experience, and has problems such as low efficiency, poor consistency, and difficulty in adapting to complex curved surface structures. Manual operation can easily cause over-grinding or under-grinding, seriously affecting the quality of repair and production efficiency. Existing automated polishing technologies are mostly based on preset paths or simple models, such as patent CN202510159660.2, which only relies on digital models to generate fixed polishing paths, resulting in improper polishing of areas with sudden changes in curvature; or like patent CN202411752959.5, the path planning and adjustment algorithms are separated, and feature matching relies on the basic Harris corner detection, which does not capture complex surface features sufficiently, resulting in difficulty in balancing processing efficiency and surface finish. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to propose a polishing method for an intelligent polishing robot that can solve the problems of the existing polishing process relying on manual experience, poor adaptability of fixed path planning, and difficulty in balancing processing efficiency and surface finish; another purpose of the present invention is to provide an intelligent polishing robot system.

[0004] Technical solution: The polishing method of the intelligent polishing robot described in the present invention comprises the following steps:

[0005] (1) Acquire three-dimensional point cloud data of the polishing target surface through a three-dimensional information acquisition device, preprocess the acquired point cloud data, extract the geometric features of the target surface from the preprocessed point cloud data, and then combine the features into a feature vector;

[0006] (2) The extracted feature vectors are classified through each decision tree, and the final classification results are determined through a voting mechanism to divide the target surface into different areas;

[0007] (3) Design adaptive grinding paths for different areas;

[0008] (4) Automated grinding: controlling the robotic arm and grinding equipment to perform grinding operations according to the generated grinding path.

[0009] Furthermore, the point cloud data preprocessed in step (1) includes denoising, downsampling and registration operations.

[0010] Furthermore, the geometric features in step (1) include normal direction, curvature, point cloud density and height difference.

[0011] Furthermore, the implementation process of step (3) is: first determine the area type, by calculating the maximum principal curvature and the minimum principal curvature of each point in the point cloud. If the difference between the two is greater than the set threshold, it is determined to be a free surface, otherwise it is a composite surface. The grinding path is adaptively generated for the free surface and the composite surface.

[0012] Furthermore, the composite curved surface includes holes, convex surfaces, concave surfaces, and plane surfaces.

[0013] Furthermore, the implementation process of determining the area type is as follows:

[0014] (31) For each point p in the point cloud i , calculate its maximum principal curvature k1 and minimum principal curvature k2, and obtain curvature information through eigenvalue decomposition of the covariance matrix C:

[0015]

[0016] Where, is the centroid of the neighborhood point set, k is the number of neighborhood points; the covariance matrix C is decomposed into eigenvalues ​​to obtain eigenvalues ​​λ1, λ2, λ3 (sorted from large to small), and the principal curvature is calculated:

[0017]

[0018] (32) If point p i If the difference between the maximum principal curvature k1 and the minimum principal curvature k2 is greater than the set threshold τ, the point is determined to belong to a free surface. For a non-free surface, it is determined to be a composite surface, and its composite surface composition is further determined;

[0019] If satisfied:

[0020] ρ i <ρ th And Δh i >Δh hole

[0021] It is determined to be a hole;

[0022] Where, ρ i is the point cloud density, ρ th is the point cloud density threshold, height difference threshold h hole =μ h +2σ h , μ h and σ h are the target surface average height difference and its standard deviation respectively;

[0023] If satisfied:

[0024] k i <-k thAnd Δh i >Δh convex

[0025] It is determined to be a concave area;

[0026] If satisfied:

[0027] k i >k th And Δh i >Δh convex

[0028] It is determined to be a convex area;

[0029] Where: k i is the point cloud curvature, k th is the curvature threshold, height difference threshold h convex =μ h +σ h ;

[0030] If satisfied:

[0031] |k i |≤k flat And Δh i ≤Δh flat

[0032] It is determined to be a flat area;

[0033] k flat is the curvature threshold, height difference threshold h flat =μ h -σ h .

[0034] Furthermore, the step of generating the grinding path includes:

[0035] Free-form surface grinding path generation: Generate isoparametric lines as the initial path based on curvature distribution, adjust the path spacing by curvature adaptation, and adjust the path spacing by curvature adaptation Where α is the process parameter, k avg is the mean curvature;

[0036] Composite surface grinding path generation: For hole areas and concave / convex areas, generate a spiral path of the boundary, the spiral path radius R k As the number of layers k decreases, the formula is:

[0037] R k =R0-k·d, d=βR0

[0038] Where R0 is the initial radius and β is the proportional coefficient;

[0039] The height difference between the layers of the path depends on the vertical drop Δz of each layer of the path satisfying:

[0040]

[0041] Where h total is the regional depth, N layer is the total number of layers;

[0042] For a plane region, equidistant section lines are generated along the long axis of the plane region. The line spacing d is fixed, and the section line equation is defined as:

[0043]

[0044] Where, L y is the length of the plane area along the y-axis, and y0 is the starting coordinate;

[0045] B-spline curve fitting is performed on discrete points on the cross-section line to ensure that the path is continuous and differentiable; transition arcs are inserted between adjacent cross-section lines to optimize the robot arm's travel path.

[0046] Furthermore, the intelligent polishing robot system described in the present invention includes a three-dimensional information acquisition device: used to collect three-dimensional point cloud data of the polishing target surface; a robotic arm: used to perform auxiliary information acquisition and polishing operations, with multi-degree-of-freedom movement capabilities, and can accurately position and move according to the polishing path; a polishing device: installed at the end of the robotic arm, used to polish the target surface.

[0047] Furthermore, the three-dimensional point cloud data includes the geometric shape, size and surface features of the target.

[0048] Furthermore, a storage medium of a computer executable program is provided, wherein the computer executable program implements the polishing method of the intelligent polishing robot when executed by a computer processor.

[0049] The beneficial effects of the present invention are: the present invention adaptively generates a polishing path through point cloud processing and machine learning, which is compared with the existing method of generating a polishing path based on a preset path or a simple geometric model; by dynamically sensing the geometric features of the target surface and generating a curvature-driven adaptive polishing path in real time, the dynamic perception and adaptive adjustment capabilities of the target surface features are improved, and the curvature mutation area is adapted. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is an operational flow chart of the present invention. DETAILED DESCRIPTION

[0051] The specific technical solutions of the present invention are further described in detail below with reference to specific examples.

[0052] As shown in the figure, the polishing method of the intelligent polishing robot described in the present invention has the following operating steps:

[0053] (1) Acquire 3D point cloud data of the polishing target surface through 3D information acquisition equipment, preprocess the acquired point cloud data, including denoising, downsampling and registration operations, extract the geometric features of the target surface from the preprocessed point cloud data, including normal direction, curvature, point cloud density and height difference, and combine these features into a feature vector;

[0054] (2) The extracted feature vectors are classified through each decision tree, and the final classification results are determined through a voting mechanism to divide the target surface into different areas;

[0055] (3) Designing adaptive polishing paths for different areas: First, determine the area type by calculating the maximum and minimum principal curvatures of each point in the point cloud. If the difference between the two is greater than a set threshold, it is determined to be a free-form surface; otherwise, it is a composite surface. The composite surface includes holes, convex surfaces, concave surfaces, and planes. Adaptive polishing paths are generated for free-form surfaces and composite surfaces.

[0056] (4) Automated grinding: controlling the robotic arm and grinding equipment to perform grinding operations according to the generated grinding path.

[0057] The processing process of step (1) is as follows:

[0058] (11) Select a binocular structured light camera to obtain the polishing target point cloud information;

[0059] (12) For the collected point cloud data, statistical filtering is used to filter outliers, and radius filtering is used to filter noise points; for downsampling of the data, for areas with large curvature changes, a smaller voxel side length is used to retain details, and for flat areas, a larger voxel side length is used to reduce the amount of data, and the point cloud space is divided into a uniform voxel grid. For the point cloud data in each voxel, its centroid point is retained and other points are removed, thereby reducing the amount of data; for the point cloud data scanned from multiple perspectives, an iterative nearest point algorithm is used for registration, and the multi-perspective point clouds are aligned to a unified coordinate system by minimizing the distance error between the point clouds;

[0060] (13) Extracting and constructing feature vectors. The implementation process includes:

[0061] (131) Calculate the normal direction of the point cloud. For each point p in the point cloud, i , select the neighborhood point set N(p i ), calculate the neighborhood point set N(p i ) of the covariance matrix C, perform eigenvalue decomposition on the covariance matrix C, and obtain the eigenvalues ​​λ1≤λ2≤λ3 and the corresponding eigenvectors v1, v2, v3. The eigenvector v1 corresponding to the minimum eigenvalue λ1 is the point p i Normal direction ni ;

[0062] (132) Use the eigenvalues ​​of the covariance matrix C to calculate the point p i The curvature k i , and perform local smoothing on the curvature value, using the average curvature of the neighborhood point set as the smoothed curvature value to reduce the impact of noise;

[0063] (133) For each point p in the point cloud i , according to the number of neighborhood points N within its radius r r (p i ) Calculate the point cloud density ρ i ;

[0064] (134), for each point p in the point cloud i , calculate the height difference Δh between it and the neighboring points i , the calculation formula is:

[0065]

[0066] Where z i and z j They are point p i and p j The height value of

[0067] (135) The normal direction n i , curvature k i , point cloud density ρ i and height difference Δh i Combined into feature vector f i , the formula is:

[0068] f i =[n i ,k i ,ρ i ,Δh i ].

[0069] The step (2) is to classify the feature vectors by using a random forest model to segment different areas of the target surface. The implementation process is:

[0070] (21) Generate label data for feature vectors based on geometric features. The generation rules are as follows:

[0071] Use K-means clustering to cluster the normal directions, dividing points with similar normal directions into the same area. The number of clusters is dynamically adjusted according to the target surface complexity and polishing requirements;

[0072] The point cloud is divided into high curvature areas and low curvature areas according to the curvature value. The curvature threshold can be dynamically adjusted according to the target surface complexity and polishing requirements.

[0073] The point cloud is divided into undulating areas and flat areas based on the height difference. The height difference threshold is dynamically adjusted according to the target surface complexity and polishing requirements.

[0074] The point cloud is divided into high-density areas and low-density areas according to the density value. The density threshold is dynamically adjusted according to the target surface complexity and polishing requirements;

[0075] For each point p i , whose region label y i Determined by the following function:

[0076]

[0077] (22) Initialize the random forest, set the number of decision trees T and the maximum depth and minimum number of sample splits as hyperparameters, and randomly select a feature subset for training for each decision tree;

[0078] (23) For each decision tree, the feature space is recursively partitioned, and the optimal partitioning features and partitioning points are selected to maximize the information gain. The information gain IG is calculated according to the following formula:

[0079]

[0080] Where D is the sample set of the current node, f is the segmentation feature, and D v is a subset of feature f with value v, and H(D) is the entropy of sample set D:

[0081]

[0082] Where p(y) is the proportion of category y in the sample set D;

[0083] (24) The feature vector with segmentation Input the trained random forest, for each decision tree, recursively traverse the tree structure according to the value of the feature vector until it reaches the leaf node, and obtain the region label prediction result. For each feature vector f i , count the prediction results of all decision trees, and select the region label with the most votes as the final result y i .

[0084] Step (3) is based on the regional division results of step (2), and different regions are classified into two categories: free-form surfaces and composite surfaces. Composite surfaces include holes, concave / convex surfaces, and flat surfaces. The classification rules are as follows:

[0085] For each point p in the point cloud i , calculate its maximum principal curvature k1 and minimum principal curvature k2, and obtain curvature information through eigenvalue decomposition of the covariance matrix C:

[0086]

[0087] Where, is the centroid of the neighborhood point set, k is the number of neighborhood points; the covariance matrix C is decomposed into eigenvalues ​​to obtain eigenvalues ​​λ1, λ2, λ3 (sorted from large to small), and the principal curvature is calculated:

[0088]

[0089] If point p i If the difference between the maximum principal curvature k1 and the minimum principal curvature k2 is greater than the set threshold τ, the point is determined to belong to a free surface. For a non-free surface, it is determined to be a composite surface, and its composite surface composition is further determined;

[0090] If satisfied:

[0091] ρ i <ρ th And Δh i >Δh hole

[0092] It is determined to be a hole;

[0093] Where p i is the point cloud density, ρ th is the point cloud density threshold, height difference threshold h hole =μ h +2σ h , μ h and σ h are the target surface average height difference and its standard deviation respectively;

[0094] If satisfied:

[0095] k i <-k th And Δh i >Δh convex

[0096] It is determined to be a concave area;

[0097] If satisfied:

[0098] k i >k th And Δh i >Δh convex

[0099] It is determined to be a convex area;

[0100] Where k i is the point cloud curvature, k th is the curvature threshold, height difference threshold h convex =μh +σ h ;

[0101] If satisfied:

[0102] |k i |≤k flat And Δh i ≤Δh flat

[0103] It is determined to be a flat area;

[0104] k flat is the curvature threshold, height difference threshold h flat =μ h -σ h ;

[0105] The implementation process of the method for generating grinding paths for different types of surfaces in step (3) includes:

[0106] Free-form surface grinding path generation: Generate isoparametric lines as the initial path based on curvature distribution, adjust the path spacing by curvature adaptation, and adjust the path spacing by curvature adaptation Where α is the process parameter, k avg is the mean curvature;

[0107] Composite surface grinding path generation: For hole areas and concave / convex areas, generate a spiral path of the boundary, the spiral path radius R k As the number of layers k decreases, the formula is:

[0108] R k =R0-k·d, d=βR0

[0109] Where R0 is the initial radius and β is the proportional coefficient;

[0110] The height difference between the layers of the path depends on the vertical drop Δz of each layer of the path satisfying:

[0111]

[0112] Where h total is the regional depth, N layer is the total number of layers;

[0113] For a plane region, equidistant section lines are generated along the long axis of the plane region. The line spacing d is fixed, and the section line equation is defined as:

[0114]

[0115] Where, L y is the length of the plane area along the y-axis, and y0 is the starting coordinate;

[0116] The initial path is smoothed using a B-spline curve to ensure that the path is continuous and differentiable. The control points of the B-spline curve are determined by minimizing the change in path curvature:

[0117]

[0118] Where s is the path arc length parameter;

[0119] Insert transition arcs between adjacent paths to ensure smooth motion paths. The transition arc radius r satisfies:

[0120]

[0121] Where θ is the angle between adjacent paths.

[0122] An intelligent polishing robot system includes a three-dimensional information acquisition device: used to collect three-dimensional point cloud data of the polishing target surface, including the target's geometric shape, size and surface characteristics;

[0123] Robotic arm: used to assist in information collection and polishing operations, with multi-degree-of-freedom motion capabilities, and can accurately position and move according to the polishing path;

[0124] Grinding equipment: installed at the end of the robotic arm, used to grind the target surface.

[0125] A storage medium comprising a computer executable program for performing the method as described above when the computer executable program is executed by a computer processor.

Claims

1. A polishing method for an intelligent polishing robot, characterized in that: The following steps are involved: (1) Acquire three-dimensional point cloud data of the polishing target surface through a three-dimensional information acquisition device, preprocess the acquired point cloud data, extract the geometric features of the target surface from the preprocessed point cloud data, and then combine the features into a feature vector; (2) The extracted feature vectors are classified through each decision tree, and the final classification results are determined through a voting mechanism to divide the target surface into different areas; (3) Design adaptive grinding paths for different areas; (4) Automated grinding: controlling the robotic arm and grinding equipment to perform grinding operations according to the generated grinding path.

2. The polishing method of the intelligent polishing robot according to claim 1, characterized in that: The point cloud data preprocessing in step (1) includes denoising, downsampling and registration operations.

3. The polishing method of the intelligent polishing robot according to claim 1, characterized in that: The geometric features in step (1) include normal direction, curvature, point cloud density and height difference.

4. The polishing method of the intelligent polishing robot according to claim 1, characterized in that: The implementation process of step (3) is as follows: first determine the region type, calculate the maximum principal curvature and the minimum principal curvature of each point in the point cloud, and if the difference between the two is greater than a set threshold, it is determined to be a free surface, otherwise it is a composite surface.

5. The polishing method of the intelligent polishing robot according to claim 4, characterized in that: The composite curved surface contains holes, convex surfaces, concave surfaces and flat surfaces.

6. The polishing method of the intelligent polishing robot according to claim 4, characterized in that: The implementation process of determining the area type is as follows: (31) For each point p in the point cloud i , calculate its maximum principal curvature k1 and minimum principal curvature k2, and obtain curvature information through eigenvalue decomposition of the covariance matrix C: Where, is the centroid of the neighborhood point set, k is the number of neighborhood points; the covariance matrix C is decomposed into eigenvalues ​​to obtain eigenvalues ​​λ1, λ2, λ3, and the principal curvature is calculated: (32) If point p i If the difference between the maximum principal curvature k1 and the minimum principal curvature k2 is greater than the set threshold τ, the point is judged to belong to a free surface. For a non-free surface, it is judged to be a composite surface, and its composite surface composition is determined; If satisfied: ρ i <ρ th and Δh i > Δh hole It is determined to be a hole; Where, ρ i is the point cloud density, ρ th is the point cloud density threshold, height difference threshold h hole =μ h +2σ h , μ h and σ h are the target surface average height difference and its standard deviation respectively; If satisfied: k i <-k th and Δh i > Δh convex It is determined to be a concave area; If satisfied: k i >k th and Δh i >Δh convex It is determined to be a convex area; Where: k i is the point cloud curvature, k th is the curvature threshold, height difference threshold h convex =μ h +σ h ; If satisfied: |k i |≤k flat And Δh i ≤Δh flat It is determined to be a flat area; k flat is the curvature threshold, height difference threshold h flat =μ h -σ h .

7. The polishing method of the intelligent polishing robot according to claim 4, characterized in that: The step of generating the grinding path includes: Free-form surface grinding path generation: Generate isoparametric lines as the initial path based on curvature distribution, adjust the path spacing by curvature adaptation, and adjust the path spacing by curvature adaptation Where α is the process parameter, k avg is the mean curvature; Composite surface grinding path generation: For hole areas and concave / convex areas, generate a spiral path of the boundary, the spiral path radius R k As the number of layers k decreases, the formula is: R k =R0-k·d,d=βR0 Where R0 is the initial radius and β is the proportional coefficient; The height difference between the layers of the path depends on the vertical drop Δz of each layer of the path satisfying: Where h total is the regional depth, N layer is the total number of layers; For a plane region, equidistant section lines are generated along the long axis of the plane region. The line spacing d is fixed, and the section line equation is defined as: Where, L y is the length of the plane area along the y-axis, and y0 is the starting coordinate.

8. An intelligent polishing robot system, characterized in that: including a three-dimensional information acquisition device for acquiring three-dimensional point cloud data of a polishing target surface; A robotic arm with multi-degree-of-freedom motion capabilities that assists in information collection and polishing operations; A grinding device installed at the end of a robotic arm to grind the target surface.

9. The intelligent polishing robot system according to claim 8, characterized in that: The three-dimensional point cloud data includes the geometric shape, size and surface features of the target.

10. A storage medium for a computer executable program, characterized in that: When the computer program is executed by a processor, the polishing method of the intelligent polishing robot according to any one of claims 1 to 6 is implemented.

Images