Workpiece polishing method and automatic polishing robot based on visual perception

Abstract

This invention provides a workpiece grinding method and an automated grinding robot based on vision perception. The workpiece grinding method includes the following steps: acquiring a 3D point cloud of the workpiece to be ground; matching the 3D point cloud with a standard 3D model point cloud corresponding to the workpiece; calculating the spatial deviation between the 3D point cloud and the standard 3D model point cloud, and determining the point cloud area exceeding a preset value in the spatial deviation as the grinding area; and controlling the grinding robot to grind the grinding area. The above-mentioned workpiece grinding method and automated grinding robot based on vision perception acquire the 3D point cloud of the workpiece to be ground through vision perception, calculate the spatial deviation between the two after matching the 3D point cloud and the standard model point cloud, and determine the point cloud area exceeding the preset value in the spatial deviation as the grinding area, thereby improving grinding efficiency and having good versatility and engineering application value.

Description

Technical Field

[0001] This disclosure pertains to the field of grinding equipment, and in particular to a workpiece grinding method based on vision perception and an automated grinding robot. Background Technology

[0002] With the development of intelligent manufacturing technology, robotic automatic grinding has been gradually applied in fields such as aerospace, automobile manufacturing, and equipment manufacturing. Compared with manual grinding, robotic grinding has significant advantages in terms of processing consistency, work efficiency, and labor intensity. However, in actual production processes, workpieces often have problems such as dimensional deviations, clamping errors, and inconsistent distribution of surface defects, which places higher demands on the perception and adaptive capabilities of the grinding system.

[0003] To automate and intelligentize the grinding process, it is necessary to accurately perceive the surface condition of the workpiece before grinding and generate corresponding grinding trajectories based on the actual defect locations. Therefore, integrating visual perception technology with robot control technology to construct an automated grinding system with environmental perception and autonomous decision-making capabilities has become an important research direction in this field.

[0004] Existing robotic automated grinding technologies mainly include manual teaching grinding and automated grinding methods based on offline programming. Manual teaching relies on operator experience, resulting in low efficiency and poor repeatability. While offline programming can improve efficiency, it has high requirements for the consistency of workpiece dimensions, shape, and clamping positions, making it difficult to adapt to individual differences in actual production. To enhance system flexibility, some existing technologies introduce vision sensors to inspect the workpiece surface, identifying defects through image processing or point cloud analysis. However, most vision inspection solutions are primarily used for defect detection or quality assessment, and their detection results are usually in the form of image coordinates or sensor coordinates, failing to establish a precise correspondence with the robot's coordinate system, making it difficult to directly guide the robot in grinding operations. Summary of the Invention

[0005] In view of the above, it is necessary to provide a workpiece grinding method and an automatic grinding robot based on vision perception, which uses vision perception to determine the grinding area.

[0006] This invention first provides a workpiece polishing method based on visual perception, which includes the following steps: Obtain the 3D point cloud of the workpiece to be polished; Match the 3D point cloud with the corresponding standard 3D model point cloud of the workpiece; Calculate the spatial deviation between the 3D point cloud and the standard 3D model point cloud, and determine the point cloud area with a spatial deviation exceeding a preset value as the polishing area; Control the grinding robot to grind the grinding area.

[0007] According to the vision-based workpiece polishing method, obtaining the three-dimensional point cloud of the workpiece to be polished includes: scanning the workpiece to be polished by a vision sensor installed at the end of the polishing robotic arm of the polishing robot to obtain the three-dimensional point cloud.

[0008] Before acquiring the 3D point cloud of the workpiece to be polished, a hand-eye calibration step is also included, which includes:

[0009] The robotic arm, equipped with the aforementioned vision sensor, scans a standard sphere placed in the workspace from N different poses, recording the rotation matrix of the robotic arm's end effector in the robot coordinate system at each pose. Translation vector The coordinates of the center of the sphere in the visual sensor coordinate system obtained by spherical fitting of the standard sphere. ;

[0010] Optimize the following objective function to determine the rotation matrix. Translation vector : ;

[0011] Where R is the rotation matrix from the vision sensor coordinate system to the robot end effector coordinate system, and t is the corresponding translation vector. Let be the coordinates of the center of the standard sphere in the visual sensor coordinate system during the i-th measurement; This represents the average position of all sphere centers in the robot's base coordinate system. The rotation matrix represents the number of robot poses collected. For the optimal value of parameter R, the translation vector The optimal value for parameter t;

[0012] According to the rotation matrix Translation vector The coordinates of the center of the sphere in the visual sensor coordinate system Coordinates converted to robot coordinate system ;

[0013] If the coordinates of the center of the sphere in the robot coordinate system If the deviation from the standard value is less than the preset value, then the rotation matrix is ​​determined. Translation vector .

[0014] According to the vision-based workpiece grinding method, obtaining the three-dimensional point cloud of the workpiece to be ground includes:

[0015] The two-dimensional contour data of the scanned workpiece with grinding is processed by the rotation matrix. Translation vector The point cloud is converted to a robot coordinate system;

[0016] The three-dimensional point cloud is sampled and filtered using a voxel grid filter to obtain the measured point cloud. .

[0017] According to the vision-based workpiece polishing method, matching the 3D point cloud with the standard 3D model point cloud corresponding to the workpiece includes:

[0018] For the measured point cloud Sample the measured point cloud and the standard point cloud, respectively, and calculate the measured point cloud. and the feature histogram features of standard point clouds;

[0019] Based on the measured point cloud The measured point cloud is matched with the feature histogram features of the standard point cloud. And standard point cloud, and obtain the initial pose transformation matrix. ;

[0020] N key structural features are obtained from the matched measured point cloud. and the key structural features corresponding to the standard point cloud. Alignment;

[0021] The following objective function is optimized to obtain the pose correction transformation matrix. : ;

[0022] in, This is the nth key structural feature of the measured point cloud. This is the nth key structural feature of the standard point cloud; The weight corresponding to the nth key structural feature; Let be the pose correction transformation to be solved, and let be the pose correction transformation matrix. for The optimal solution;

[0023] Calculate the pose transformation matrix :

[0024] by As the initial pose, the optimal solution of T is obtained by iteratively solving the following objective function, which serves as the pose transformation matrix. : ;

[0025] in, , For the first point cloud One point, For the corresponding point in the standard point cloud, For point The normal vector at that location; Let be the rigid body transformation matrix to be solved;

[0026] Through pose transformation matrix Align the measured point cloud with the standard model.

[0027] According to the vision-based workpiece polishing method, the spatial deviation between the three-dimensional point cloud and the standard three-dimensional model point cloud is calculated, and the point cloud areas with spatial deviations exceeding a preset value are identified as areas to be polished, including:

[0028] Calculate the i-th measured point in the measured point cloud. The surface of the model point cloud is determined by the measured points. The corresponding point with the smallest Euclidean distance The Euclidean distance between them is used as the measured point. deviation value : ;

[0029] in, It is a norm 2;

[0030] If the deviation value If the value is greater than the preset process threshold, then the measured point will be... The set of points to be polished is determined as D.

[0031] According to the vision-based workpiece grinding method, for any grinding point in the set D of points to be ground... The number of points in its neighborhood is defined as: ;

[0032] in, In the set D of points to be polished, excluding any other point besides The neighborhood radius threshold, To find the minimum number of neighborhood points, when satisfying At that time, the polishing point is determined as the core point;

[0033] By iteratively expanding the core point and its neighboring polishing points, the spatially densely distributed burr points are divided into the same defect region, forming several independent clusters of polishing points. .

[0034] According to the vision-based workpiece polishing method, controlling the polishing robot to polish the area to be polished includes:

[0035] For each polishing point cloud cluster For the polishing point cloud cluster Modeling and calculating the grinding point cloud clusters. covariance matrix : ; ;

[0036] Where, N n p represents the total number of points in the nth polishing point cloud cluster. k For the k-th point in this cluster, μ n Let the centroid of this cluster be the covariance matrix. Perform eigenvalue decomposition and take the eigenvector corresponding to the smallest eigenvalue as the normal vector of the reference plane of the polishing area. And it is defined as the feed direction for robot grinding;

[0037] Polishing point cloud clusters Each point in the matrix is ​​orthogonally projected onto the reference plane, and the set of projected points is calculated. : ;

[0038] Based on projection point set Extract the two-dimensional contour and generate a two-dimensional mesh lattice covering the burr area in the reference plane based on the radius of the grinding tool. ;

[0039] Calculate the maximum height h of the burr area in the feed direction: ;

[0040] According to the preset floor height Along the normal vector Spatial stratification is performed in the opposite direction, the first The path point set of a layer is defined as follows

[0041] ; The number of layers in the burr region is represented by the i-th grid point in the two-dimensional grid.

[0042] For each layer of path point set The nearest neighbor criterion is used to sort path points. The order in which path points are visited is determined by minimizing the Euclidean distance between adjacent path points, i.e.:

[0043] ; Where is the sorted path point sequence and is the number of path points in the i-th layer.

[0044] Each layer path is connected sequentially according to the preset layer number, thus forming a top-down layered polishing trajectory.

[0045] According to the vision-based workpiece polishing method, controlling the polishing robot to polish the area to be polished further includes a step for transitioning polishing paths corresponding to different polishing areas, which includes:

[0046] The transition trajectory is configured using the following three consecutive pose points:

[0047] ;

[0048] ;

[0049] ;

[0050] in, To preset a safe lifting height, and These represent the end point of the sanding path in the previous sanding area and the start point of the next sanding path, respectively. This refers to the grinding feed direction for the corresponding grinding area;

[0051] The actuators at the end of the robotic arm pass through the pose points in sequence. pose point and pose point This enables safe movement through lifting, translation, and descent.

[0052] According to the vision-based workpiece grinding method, after configuring the transition trajectory, the method further includes:

[0053] For the three-dimensional path point sequence The trajectory expression for B-spline curve interpolation smoothing is as follows:

[0054] ;

[0055] in, For points on the B-spline curve, These are the control points for the B-spline curve. for B-order spline basis functions.

[0056] The present invention also provides an automated polishing robot, including a robotic arm and an actuator and a vision sensor disposed at the end of the robotic arm, and further comprising:

[0057] Memory that stores non-transitory computer-readable instructions;

[0058] At least one processor is configured to perform at least one step of the vision-based workpiece polishing method when executing the non-transitory computer-readable instructions.

[0059] Compared to existing technologies, the aforementioned vision-based workpiece grinding method and automatic grinding robot acquire the three-dimensional point cloud of the workpiece to be ground through visual perception. After matching the three-dimensional point cloud with the standard model point cloud, the spatial deviation between the two is calculated. The point cloud area with a spatial deviation exceeding a preset value is used as the grinding area. This not only improves grinding efficiency, but also maintains high grinding accuracy, especially in grinding various workpieces with holes or key structural features, even when the workpiece posture changes or the batch varies greatly. It has good versatility and engineering application value. Attached Figure Description

[0060] To more clearly illustrate the specific implementation methods, the accompanying drawings used in the description of the implementation methods will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 A schematic diagram of the structure of an automated grinding robot.

[0062] Figure 2 This is a flowchart of the workpiece grinding method.

[0063] Figure 3 This is a flowchart for matching 3D point clouds with model point clouds.

[0064] Figure 4 This is a schematic diagram of matching 3D point clouds with model point clouds.

[0065] Figure 5 This is a schematic diagram illustrating the extraction of key structural features during the polishing process.

[0066] Figure 6 This is a schematic diagram of the layered path point planning for a single polishing area.

[0067] Figure 7 This is a schematic diagram of the polishing path.

[0068] Explanation of component symbols:

[0069] 10 - Robotic arm; 20 - Vision sensor; 30 - Actuator; 40 - Workpiece; 50 - Standard ball.

[0070] The following detailed embodiments will further illustrate this disclosure in conjunction with the above-described drawings. Detailed Implementation

[0071] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. Many specific details are set forth in the following description to provide a thorough understanding of this disclosure; the described embodiments are merely a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0072] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure.

[0073] In various embodiments, for ease of description and not limitation of this disclosure, the term "connection" used in the patent application specification and claims is not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "below," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0074] Figure 1 A schematic diagram of the structure of an automated polishing robot. Figure 1 As shown, the automated grinding robot includes a robotic arm 10 and an actuator 30 and a vision sensor 20 disposed at the end of the robotic arm 10. For example, the vision sensor 20 can be a stereo camera or a line laser profilometer, etc., capable of recognizing the contour shape of the workpiece 40. The robotic arm 10 can be a six-axis industrial robotic arm, with the actuator 30 (i.e., the grinding component) connected to its end (i.e., the free end) to grind the workpiece 40. Both the vision sensor 20 and the actuator 30 are disposed at the end of the robotic arm 10, achieving an "eye-hand" configuration.

[0075] In addition, the automated polishing robot also includes a memory and at least one processor. The memory stores non-transitory computer-readable instructions, and the storage device can be different types of storage devices used to store various types of data. For example, it can be the automated polishing robot's RAM, or an externally connectable memory card, such as flash memory, SM card (Smart Media Card), SD card (Secure Digital Card), etc. The processor includes, but is not limited to, a Central Processing Unit (CPU), a Microcontroller Unit (MCU), or other devices used to interpret computer instructions and process data in computer software. The processor is configured to execute at least one step of the vision-based workpiece polishing method when executing the non-transitory computer-readable instructions.

[0076] Figure 2 This is a flowchart of the workpiece grinding method. For example... Figure 2 As shown, the workpiece grinding method based on visual perception includes steps S1-6. In this embodiment, the workpiece 40 is a flange with a center hole and bolt holes as the grinding object, and its typical defect is the burr or flash generated on the outer edge after machining. However, the workpiece 40 can also be other shapes, and the grinding object is not limited to the flange shape.

[0077] S1: Hand-eye calibration steps. This step first collects data from a standard sphere 50, then builds and solves the model, and finally verifies the accuracy of the calibration structure to confirm the accuracy of the hand-eye calibration. The aim is to establish a mapping between the vision sensor coordinate system and the robot coordinate system.

[0078] Specifically, data acquisition is performed first: Before polishing, a standard ball 50 of known diameter (or other structural parts of known shape) is placed in the workspace of the polishing robot. The robotic arm 10, carrying the vision sensor 20, scans the standard ball 50 placed in the workspace from N (N preferably not less than 10) different poses, and records the rotation matrix of the end effector of the robotic arm 10 in the robot coordinate system at each pose. Translation vector The coordinates of the center of the sphere in the visual sensor coordinate system obtained by spherical fitting of the standard sphere 50. .

[0079] Then, model building and solving are performed: the goal of hand-eye calibration is to solve for the rotation matrix from the vision sensor coordinate system to the robot end effector coordinate system. Translation vector In this step, the following objective function is constructed using a nonlinear least squares optimization model, and the rotation matrix is ​​determined by optimizing the objective function. Translation vector :

[0080] ;

[0081] Where R is the rotation matrix from the vision sensor coordinate system to the robot end effector coordinate system, and t is the corresponding translation vector. Let be the coordinates of the center of the standard sphere in the visual sensor coordinate system during the i-th measurement; This represents the average position of all sphere centers in the robot's base coordinate system. The rotation matrix represents the number of robot poses collected. For the optimal value of parameter R, the translation vector To find the optimal value of parameter t, the rotation matrix at the optimal value of the function is obtained by optimizing the function. Translation vector .

[0082] Finally, verify the calibration accuracy: based on the stated rotation matrix. Translation vector The coordinates of the center of the sphere in the visual sensor coordinate system Coordinates converted to robot coordinate system If the coordinates of the center of the sphere in the robot coordinate system are... If the deviation from the standard value is less than the preset value, then the rotation matrix is ​​determined. Translation vector For example, if the average deviation of all ball center positions is less than 0.1 mm and the maximum deviation is less than 0.2 mm, then the accuracy of the hand-eye calibration structure meets the preset requirements. This method avoids the error propagation problem caused by step-by-step solutions and features strong noise resistance and high calibration accuracy.

[0083] S2: Obtain the 3D point cloud of the workpiece 40 to be polished. In this step, the 3D point cloud is obtained by scanning the workpiece 40 to be polished using a vision sensor 20 installed at the end of the polishing robotic arm 10 of the polishing robot. Specifically, the scanned 2D contour data of the workpiece 40 to be polished is processed by the rotation matrix. Translation vector The point cloud is converted to a robot coordinate system, and then sampled and filtered using a voxel grid filter to remove outlier noise points, resulting in a clean and uniform measured point cloud. This provides high-quality input for subsequent processing.

[0084] S3: Match the three-dimensional point cloud with the standard three-dimensional model point cloud corresponding to the workpiece 40. Figure 3 This is a flowchart for matching 3D point clouds with model point clouds. For example... Figure 3As shown, the purpose of this step is to align the 3D point cloud with the corresponding model point cloud by first coarse registration, then optimizing the pose through feature structure, and finally fine matching.

[0085] S31: Coarse matching steps:

[0086] First, the measured point cloud Sample the measured point cloud and the standard point cloud, respectively, and calculate the measured point cloud. And the feature histogram (FPFH) features of standard point clouds.

[0087] Then, based on the measured point cloud The measured point cloud is matched with the feature histogram features of the standard point cloud. And standard point cloud, and obtain the initial pose transformation matrix. . Figure 4 This is a schematic diagram illustrating the matching of 3D point clouds with model point clouds. For example... Figure 4 As shown, a feature matching algorithm based on Random Sample Consensus (RANSAC) is used to match the measured point cloud. Match the FPFH features of the standard point cloud to solve for the initial pose transformation matrix. Through the initial pose transformation matrix Coarse Alignment Measured Point Cloud And standard point clouds to eliminate approximate pose deviations caused by clamping and placement.

[0088] S32: Feature structure optimization pose steps:

[0089] First, N key structural features are obtained from the matched measured point cloud. and the key structural features corresponding to the standard point cloud. Alignment. Key structural features are features pre-specified from standard point clouds, preferably features with well-defined geometry that satisfy one or more of the following combinations:

[0090] (1) Process reference: This feature serves as a positioning or measurement reference in the design and manufacturing of workpiece 40. It has high machining accuracy and can most realistically reflect the theoretical design position of workpiece 40.

[0091] (2) Geometric constraints: The geometric model of this feature can provide strong constraints on multiple spatial degrees of freedom of the workpiece 40. For example, the cylindrical surface along its axial direction, the center of the sphere, two intersecting planes, and the corner points of the outer contour.

[0092] (3) Point cloud robustness: In the measured point cloud, the point set constituting this feature has high integrity and low noise, and can be stably and accurately extracted through model fitting algorithm.

[0093] Figure 5 This is a schematic diagram illustrating the extraction of key structural features during the polishing process. For example... Figure 5 As shown, taking the center hole of a flange as an example: the cylindrical feature of the center hole is obtained by fitting from the measured point cloud after coarse registration, and the direction vector of its measured axis is obtained. and center point and the corresponding axis direction vector in the standard model. and center point (2) Align.

[0094] Then, the following objective function is optimized to obtain the pose correction transformation matrix. :

[0095] ;

[0096] in, This is the nth key structural feature of the measured point cloud. This is the nth key structural feature of the standard point cloud; The weight corresponding to the nth key structural feature; Let be the pose correction transformation to be solved, and let be the pose correction transformation matrix. for The optimal solution.

[0097] S33: Fine-matching steps:

[0098] Calculate the pose transformation matrix :

[0099] by As the initial pose, the point-to-surface ICP algorithm is used for fine registration, and the optimal solution of the following objective function, T, is obtained by iteratively solving the problem and serving as the pose transformation matrix. :

[0100] ;

[0101] in, , For the first point cloud One point, For the corresponding point in the standard point cloud, For point The normal vector at that location; Let be the rigid body transformation matrix to be solved;

[0102] Finally, through the pose transformation matrix Align the measured point cloud with the standard model.

[0103] S4: Calculate the spatial deviation between the 3D point cloud and the standard 3D model point cloud, and determine the point cloud regions with spatial deviations exceeding a preset value as polishing regions. In this step, based on the 3D spatial deviation, one or more independent polishing point cloud clusters are formed by segmenting and clustering the defective regions. .

[0104] Specifically, firstly, using the final pose transformation matrix After aligning the measured point cloud with the model point cloud, calculate the i-th measured point of the transformed measured point cloud. The surface of the model point cloud is determined by the measured points. The corresponding point with the smallest Euclidean distance The Euclidean distance between them is used as the measured point. deviation value :

[0105] ;

[0106] in, It is a norm 2. If the deviation value... If the measured value is greater than the preset process threshold (e.g., 0.15 mm), then the measured point will be... The set of points to be polished is determined as D.

[0107] Then, a density-based clustering algorithm is used to cluster the points to be polished, based on the set D of points to be polished. Specifically, for any polishing point in the set D of points to be polished... The number of points in its neighborhood is defined as:

[0108] ;

[0109] in, In the set D of points to be polished, excluding any other point besides The neighborhood radius threshold, To find the minimum number of neighborhood points, when satisfying At that time, the polishing point is determined as the core point.

[0110] Finally, by iteratively expanding the core point and the polishing points in its neighborhood, the spatially densely distributed polishing points are divided into the same defect region, forming several independent clusters of polishing points. .

[0111] The grinding point clusters of the workpiece 40 to be ground are determined through steps S1-4. Next, it is necessary to plan the grinding path for the grinding area. The following provides an adaptive grinding path planning method for layered cutting (i.e., step S5).

[0112] S5: Plan the polishing path for the polishing area. For each polishing point cluster... The grinding path is generated layer by layer from top to bottom based on the path planning method. Specifically, the grinding robot is controlled to grind the area to be ground by region modeling and layering, then the grinding path of each layer is optimized and connected, and finally the grinding trajectory is smoothed to generate the grinding path.

[0113] Region modeling and stratification steps:

[0114] This step uses principal component analysis for fitting. The bottom reference plane, the normal vector of the bottom reference plane Define the grinding feed direction, Projected onto the reference plane, a two-dimensional mesh is generated based on the two-dimensional contour and the radius of the grinding tool, and then along the normal vector. In the opposite direction, spatial layering is performed using a preset layer height. For each polishing point cloud cluster... Generate a grinding path that removes material layer by layer from top to bottom.

[0115] Specifically, principal component analysis was used to analyze the polishing point cloud clusters. Modeling and calculating the grinding point cloud clusters. center of mass :

[0116] ;

[0117] Then, based on the construction and polishing of the cloud clusters covariance matrix :

[0118] ;

[0119] Where, N n p represents the total number of points in the nth polishing point cloud cluster. k For the k-th point in this cluster, μ n Let the centroid of this cluster be the covariance matrix. Perform eigenvalue decomposition and take the eigenvector corresponding to the smallest eigenvalue as the normal vector of the reference plane of the polishing area. And it is defined as the feed direction for robot grinding.

[0120] Next, the polished cloud clusters will be... Each point in the matrix is ​​orthogonally projected onto the reference plane, and the set of projected points is calculated. :

[0121] ;

[0122] Then, based on the projection point set Extract the two-dimensional contour and generate a two-dimensional mesh lattice covering the burr area in the reference plane based on the radius of the grinding tool. .

[0123] Finally, calculate the maximum height of the grinding area in the feed direction. :

[0124] ;

[0125] According to the preset floor height Along the normal vector Spatial stratification is performed in the opposite direction, the first The path point set of a layer is defined as follows

[0126] ;

[0127] in: The number of layers in the burr area. The i-th grid point in a two-dimensional grid;

[0128] Path optimization and connection steps:

[0129] In this step, the path points within each layer are sorted using the nearest neighbor algorithm, and the paths of each layer are connected in a top-down order. A safe movement strategy of lifting, translating, and descending is adopted between different polishing areas.

[0130] Figure 6 This is a schematic diagram of the layered path point planning for a single polishing area. For example... Figure 6 As shown, for each layer of path point set The nearest neighbor criterion is used to sort path points. The order in which path points are visited is determined by minimizing the Euclidean distance between adjacent path points, i.e.:

[0131] ;

[0132] in, This is the sorted sequence of path points. For the first Number of layer path points.

[0133] Then, the paths of each layer are connected sequentially according to the preset order of the layer numbers (for example, in ascending order of layer numbers), thus forming a top-down layered polishing trajectory.

[0134] For grinding paths corresponding to different grinding areas, actuator 30 adopts a safe transition strategy when switching paths, which includes the following steps for transitioning between grinding paths corresponding to different grinding areas:

[0135] The transition trajectory is configured using the following three consecutive pose points:

[0136]

[0137]

[0138] ;

[0139] in, To preset a safe lifting height, and These represent the end point of the sanding path in the previous sanding area and the start point of the next sanding path, respectively. This corresponds to the grinding feed direction of the grinding area. Thus, the actuator 30 at the end of the robotic arm 10 sequentially passes through the aforementioned pose points. pose point and pose point This enables safe movement through lifting, translation, and descent.

[0140] Trajectory smoothing steps :

[0141] Figure 7 This is a diagram illustrating the polishing path. For example... Figure 7 As shown, in this step, the final generated three-dimensional path point sequence is smoothed by B-spline curve interpolation to obtain a continuous and smooth robot polishing trajectory.

[0142] Specifically, for the three-dimensional path point sequence The trajectory expression for B-spline curve interpolation smoothing is as follows:

[0143] ;

[0144] in,, For points on the B-spline curve, These are the control points for the B-spline curve. for B-order spline basis functions.

[0145] S6: Control the polishing robot to polish the polishing area, convert the polishing trajectory data into robot motion control commands, and control the robotic arm 10 to polish along the preset trajectory to complete the polishing operation. Specifically, the robot receives data containing the polishing trajectory. The file is converted into robot motion control instructions. The end effector 30 carries the force-controlled floating grinding head and moves along the planned trajectory. At the same time, it maintains a constant contact force of about 30N through a force / position hybrid control strategy, adapting to the micro-undulations on the surface of the workpiece 40, ensuring that burrs are removed evenly and accurately without damaging the workpiece 40.

[0146] The aforementioned vision-based workpiece grinding method and automatic grinding robot acquire the three-dimensional point cloud of the workpiece 40 to be ground through vision perception. After matching the three-dimensional point cloud with the standard model point cloud, the spatial deviation between the two is calculated. The point cloud area with a spatial deviation exceeding the preset value is taken as the grinding area. This not only improves grinding efficiency, but also maintains high grinding accuracy, especially in grinding various workpieces 40 with holes or key structural features, even when the workpiece 40 posture changes or the batch varies greatly. It has good versatility and engineering application value.

[0147] Furthermore, the aforementioned vision-based workpiece grinding method and automatic grinding robot establish a spatial mapping relationship between the vision sensor coordinate system and the robot coordinate system through hand-eye calibration, thereby realizing the accurate expression of the vision perception results in the robot coordinate system and improving the spatial positioning accuracy of the grinding operation.

[0148] By introducing key structural features of workpiece 40 as geometric constraints into the matching of the three-dimensional point cloud with the corresponding three-dimensional standard model point cloud, the problem of key structural offset caused by planar features in the traditional point cloud registration method is effectively avoided, and the alignment accuracy of the overall and local structures of workpiece 40 is improved.

[0149] Moreover, based on accurate registration, by analyzing the geometric deviation between the actual point cloud of the workpiece 40 and the standard model, the automatic extraction of the defect area to be polished was realized, reducing manual intervention and improving the objectivity and consistency of defect identification. Finally, the defect extraction results were directly used for polishing path planning, realizing the effective connection between defect identification and robot motion planning, and avoiding the problem of independent detection and processing in the existing technology.

[0150] Furthermore, the aforementioned vision-based workpiece grinding method and automatic grinding robot form a complete automated process from vision perception, pose determination, defect identification to grinding path generation, which improves the overall intelligence level and operational efficiency of the robot automatic grinding system.

[0151] In the several specific embodiments provided in this disclosure, it will be apparent to those skilled in the art that this disclosure is not limited to the details of the exemplary embodiments described above, and that this disclosure can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of this disclosure is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within this disclosure. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Terms such as "first," "second," etc., are used to denote names and do not indicate any particular order.

[0152] The above embodiments are only used to illustrate the technical solutions of this disclosure and are not intended to limit it. Although this disclosure has been described in detail with reference to the above preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this disclosure should not depart from the spirit and scope of the technical solutions of this disclosure.

Claims

1. A workpiece grinding method based on visual perception, characterized in that, Includes the following steps: Obtain the 3D point cloud of the workpiece to be polished; Match the 3D point cloud with the corresponding standard 3D model point cloud of the workpiece; Calculate the spatial deviation between the 3D point cloud and the standard 3D model point cloud, and determine the point cloud area with a spatial deviation exceeding a preset value as the polishing area; Control the grinding robot to grind the grinding area.

2. The workpiece grinding method based on vision perception as described in claim 1, characterized in that, Obtaining the three-dimensional point cloud of the workpiece to be polished includes: scanning the workpiece to be polished using a vision sensor installed at the end of the polishing robotic arm of the polishing robot to obtain the three-dimensional point cloud; Before acquiring the 3D point cloud of the workpiece to be polished, a hand-eye calibration step is also included, which includes: The robotic arm, equipped with the aforementioned vision sensor, scans a standard sphere placed in the workspace from N different poses, recording the rotation matrix of the robotic arm's end effector in the robot coordinate system at each pose. Translation vector The coordinates of the center of the sphere in the visual sensor coordinate system obtained by spherical fitting of the standard sphere. ; Optimize the following objective function to determine the rotation matrix. Translation vector : ； Where R is the rotation matrix from the vision sensor coordinate system to the robot end effector coordinate system, and t is the corresponding translation vector. Let be the coordinates of the center of the standard sphere in the visual sensor coordinate system during the i-th measurement; This represents the average position of all sphere centers in the robot's base coordinate system. The rotation matrix represents the number of robot poses collected. For the optimal value of parameter R, the translation vector The optimal value for parameter t; According to the rotation matrix Translation vector The coordinates of the center of the sphere in the visual sensor coordinate system Coordinates converted to robot coordinate system ; If the coordinates of the center of the sphere in the robot coordinate system If the deviation from the standard value is less than the preset value, then the rotation matrix is ​​determined. Translation vector .

3. The workpiece grinding method based on vision perception as described in claim 2, characterized in that, Obtaining the 3D point cloud of the workpiece to be polished includes: The two-dimensional contour data of the scanned workpiece with grinding is processed by the rotation matrix. Translation vector The point cloud is converted to a robot coordinate system; The three-dimensional point cloud is sampled and filtered using a voxel grid filter to obtain the measured point cloud. .

4. The workpiece grinding method based on vision perception as described in claim 3, characterized in that, Matching the 3D point cloud with the corresponding standard 3D model point cloud of the workpiece includes: For the measured point cloud Sample the measured point cloud and the standard point cloud, respectively, and calculate the measured point cloud. and the feature histogram features of standard point clouds; Based on the measured point cloud The measured point cloud is matched with the feature histogram features of the standard point cloud. And standard point cloud, and obtain the initial pose transformation matrix. ; N key structural features are obtained from the matched measured point cloud. and the key structural features corresponding to the standard point cloud. Alignment; The following objective function is optimized to obtain the pose correction transformation matrix. : ； in, This is the nth key structural feature of the measured point cloud. This is the nth key structural feature of the standard point cloud; The weight corresponding to the nth key structural feature; Let be the pose correction transformation to be solved, and let be the pose correction transformation matrix. for The optimal solution; Calculate the pose transformation matrix : by As the initial pose, the optimal solution of T is obtained by iteratively solving the following objective function, which serves as the pose transformation matrix. : ; in, , For the first point cloud One point, For the corresponding point in the standard point cloud, For point The normal vector at that location; Let be the rigid body transformation matrix to be solved; Through pose transformation matrix Align the measured point cloud with the standard model.

5. The workpiece grinding method based on vision perception as described in claim 4, characterized in that, Calculating the spatial deviation between the 3D point cloud and the standard 3D model point cloud, and determining the point cloud region exceeding a preset value in the spatial deviation as the region to be polished includes: Calculate the i-th measured point in the measured point cloud. The surface of the model point cloud is determined by the measured points. The corresponding point with the smallest Euclidean distance The Euclidean distance between them is used as the measured point. deviation value : ； in, It is a norm 2; If the deviation value If the value is greater than the preset process threshold, then the measured point will be... The set of points to be polished is determined as D.

6. The workpiece grinding method based on vision perception as described in claim 5, characterized in that, For any grinding point in the set D of points to be ground The number of points in its neighborhood is defined as: ； in, In the set D of points to be polished, excluding any other point besides The neighborhood radius threshold, To find the minimum number of neighborhood points, when satisfying At that time, the polishing point is determined as the core point; By iteratively expanding the core point and its neighboring polishing points, the spatially densely distributed burr points are divided into the same defect region, forming several independent clusters of polishing points. .

7. The workpiece grinding method based on vision perception as described in claim 6, characterized in that, Controlling the polishing robot to polish the area to be polished includes: For each polishing point cloud cluster For the polishing point cloud cluster Modeling and calculating the grinding point cloud clusters. covariance matrix : ； ； Where, N n p represents the total number of points in the nth polishing point cloud cluster. k For the k-th point in this cluster, μ n Let the centroid of this cluster be the covariance matrix. Perform eigenvalue decomposition and take the eigenvector corresponding to the smallest eigenvalue as the normal vector of the reference plane of the polishing area. And it is defined as the feed direction for robot grinding; Polishing point cloud clusters Each point in the matrix is ​​orthogonally projected onto the reference plane, and the set of projected points is calculated. : ； Based on projection point set Extract the two-dimensional contour and generate a two-dimensional mesh lattice covering the burr area in the reference plane based on the radius of the grinding tool. ; Calculate the maximum height of the burr area in the feed direction. : ； According to the preset floor height Along the normal vector Spatial stratification is performed in the opposite direction, the first The path point set of a layer is defined as follows ； in: The number of layers in the burr area. Let i be the i-th grid point in the two-dimensional grid. For each layer of path point set The nearest neighbor criterion is used to sort path points. The order in which path points are visited is determined by minimizing the Euclidean distance between adjacent path points, i.e.: ； in, This is the sorted sequence of path points. For the first Number of path points in the layer; Each layer path is connected sequentially according to the preset layer number, thus forming a top-down layered polishing trajectory.

8. The workpiece grinding method based on vision perception as described in claim 7, characterized in that, Controlling the polishing robot to polish the area to be polished also includes a step for transitioning polishing paths corresponding to different polishing areas, which includes: The transition trajectory is configured using the following three consecutive pose points: ； ； ; in, To preset a safe lifting height, and These represent the end point of the sanding path in the previous sanding area and the start point of the next sanding path, respectively. This refers to the grinding feed direction for the corresponding grinding area; The actuators at the end of the robotic arm pass through the pose points in sequence. pose point and pose point This enables safe movement through lifting, translation, and descent.

9. The workpiece grinding method based on vision perception as described in claim 8, characterized in that, After configuring the transition trajectory, the following is also included: For the three-dimensional path point sequence The trajectory expression for B-spline curve interpolation smoothing is as follows: ； in, For points on the B-spline curve, These are the control points for the B-spline curve. for B-order spline basis functions.

10. An automated polishing robot, comprising a robotic arm and an actuator and a vision sensor disposed at the end of the robotic arm, characterized in that, include: Memory that stores non-transitory computer-readable instructions; At least one processor is configured to, when executing the non-transitory computer-readable instructions, perform at least one step of the vision-based workpiece polishing method according to any one of claims 1-9.

Images