A trajectory generation method, device and medium for deburring cast blanks

Abstract

The present invention discloses a trajectory generation method, device and medium for deburring cast blanks, which relates to the technical field of grinding processes. The method mainly includes the steps of: according to the FPFH feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model respectively, and searching and matching the key points of the standard point cloud model through a KD tree, dividing the multi-contours of the actual point cloud model according to the distances between the key points; receiving the polishing parameters input by the user based on the results of each contour segmentation, and generating an initial trajectory according to the polishing parameters; smoothing the discrete trajectory points of the initial trajectory through B-spline interpolation, and generating a local obstacle avoidance path in combination with the industrial robot kinematic model and the rapidly-exploring random tree star algorithm; outputting the final polishing trajectory to the control terminal, and adjusting the trajectory depth in real time according to the feedback of the force sensor. The present invention supports region-differentiated trajectory generation, and improves the process adaptability to complex contours and the ability to respond to personalized requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of grinding processes, and particularly to a method, device, and medium for generating trajectories for deburring cast blanks. Background Art

[0002] In the field of industrial automation, 3D vision-based deburring technology for castings has been gradually applied to actual production. Existing technologies usually obtain the point cloud data of castings through 3D vision imaging, register it with a pre-stored standard 3D model, and directly generate grinding trajectories based on the outer contour of the standard model. However, such methods have significant drawbacks: First, due to the limitations of the casting process or deviations during production, there are often local differences between the actual shape of the cast blank and the standard model. If the contour of the standard model is directly applied for grinding, it may result in excessive grinding depth, causing damage to the surface of the casting or even the grinding device, affecting processing safety and the qualified rate of finished products. Second, existing technologies lack adaptability to diverse requirements. For example, different processing areas have personalized requirements for parameters such as chamfer depth and lateral expansion width. Some areas need to avoid the machined surface or adopt specific feed trajectories (such as quadratic function parameterized trajectories). However, existing methods only generate fixed trajectories based on the outer contour and cannot achieve segmented parameter adjustment, resulting in limited application scenarios. In addition, existing technologies do not fully consider the dynamic registration accuracy between the actual point cloud and the standard model, and the interference of mis-matched point clouds may further reduce the reliability of trajectory generation. Therefore, there is an urgent need for a trajectory generation method that can combine the actual morphology of castings, support segmented parameter configuration, and have high robustness to improve grinding accuracy and process flexibility. Summary of the Invention

[0003] Aiming at the deficiencies of the existing technology, the present invention proposes a method for generating trajectories for deburring cast blanks, including the steps of:

[0004] S1: Scanning the cast blank with a 3D line array camera to obtain an actual point cloud model, and extracting the pre-entered standard point cloud model;

[0005] S2: Based on the improved iterative closest point algorithm, performing registration adjustment of the actual point cloud model and the standard point cloud model under coordinate system alignment;

[0006] S3: According to the FPFH feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model respectively, and searching and matching the key points of the standard point cloud model through a KD tree, and performing multi-contour segmentation of the actual point cloud model according to the distances between the key points;

[0007] S4: Receiving the grinding parameters input by the user based on the results of each contour segmentation, and generating an initial trajectory according to the grinding parameters;

[0008] S5: Smooth the discrete trajectory points of the initial trajectory through B-spline interpolation, and combine the industrial robot kinematic model and the rapidly-exploring random tree star algorithm to generate a local obstacle avoidance path;

[0009] S6: Output the final polishing trajectory to the control terminal, and adjust the trajectory depth in real time according to the feedback of the force sensor.

[0010] Further, in the step S2, the improved iterative closest point algorithm is specifically:

[0011] S21: Extract the FPFH feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model respectively, and screen the initial paired points;

[0012] S22: Eliminate the mismatched points through the random sample consensus algorithm, and perform the registration adjustment of the actual point cloud model and the standard point cloud model under the coordinate system alignment through the following formula:

[0013]

[0014] In the formula, is the rotation matrix, is the translation vector, is the total number of three-dimensional points in the point cloud model, is the th three-dimensional point in the actual point cloud model, is the th three-dimensional point in the standard point cloud model;

[0015] S23: Iteratively optimize the rotation matrix and the translation vector , until the error of the registration adjustment is less than the preset error threshold.

[0016] Further, in the step S3, the key points for searching and matching the standard point cloud model through the KD tree are expressed as the following formula:

[0017]

[0018] In the formula, is the FPFH feature descriptor of the th point in the actual point cloud model, is the th three-dimensional point in the standard point cloud model, is the th point in the standard point cloud model, is the index that finds the smallest Euclidean distance from among all the standard point cloud candidate points.

[0019] Further, in the step S4, the grinding parameters include the grinding depth , the lateral expansion width and the quadratic function feed angle of the chamfering required area , where is the coefficient input by the user, is the final grinding trajectory.

[0020] Further, the initial trajectory is expressed by the following formula:

[0021]

[0022] In the formula, is the normal offset trajectory, is the lateral expansion trajectory, is the reference trajectory obtained based on the standard point cloud model, N is the point cloud normal direction, and T is the edge tangent direction.

[0023] Further, in the step S5, the smoothing of the discrete trajectory points of the initial trajectory by B-spline interpolation is expressed by the following formula:

[0024]

[0025] In the formula, is the parameter variable, is the optimized initial trajectory, is the number of B-spline basis functions of order is the control point dynamically adjusted according to the simulation result of the industrial robot kinematic model in the initial trajectory, is the total number of control points.

[0026] Further, in the step S5, the generation of the local obstacle avoidance path by combining the industrial robot kinematic model and the RRT* algorithm is specifically as follows:

[0027] Construct the geometric bounding box model of the robot, fixture and workpiece;

[0028] Verify the feasibility of the trajectory pose through inverse kinematics;

[0029] If interference is detected, use the rapidly-exploring random tree star algorithm to generate a local obstacle avoidance path.

[0030] Further, in the step S6, the real-time adjustment of the trajectory depth according to the feedback of the force sensor is expressed by the following formula:

[0031]

[0032] In the formula, is the adjusted trajectory depth, is the original trajectory depth, is the proportionality coefficient, is the reference pressure, is the real-time pressure of the force sensor.

[0033] The present invention further includes a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the method for generating a trajectory for deburring a casting blank are implemented.

[0034] It further includes a device for processing data, including:

[0035] A memory, on which a computer program is stored;

[0036] A processor, configured to execute the computer program in the memory to implement the steps of the method for generating a trajectory for deburring a casting blank.

[0037] Compared with the prior art, the present invention has at least the following beneficial effects:

[0038] (1) The method for generating a trajectory for deburring a casting blank proposed by the present invention realizes the matching and segmentation of key points on the outer contour of the casting based on FPFH features and KD tree search. Combining with the grinding parameters input by the user, it supports the generation of differentiated trajectories in different regions (such as chamfers and burr concentration areas), significantly improving the process adaptability to complex contours and the ability to respond to personalized requirements;

[0039] (2) By combining the improved iterative closest point algorithm (ICP) with the screening of FPFH feature descriptors and the rejection of mismatches by random sample consensus (RANSAC), the registration accuracy between the actual point cloud and the standard model is significantly improved, effectively reducing the coordinate system alignment error caused by casting deformation, and avoiding over-grinding or equipment collision caused by model deviation;

[0040] (3) Through B-spline interpolation, the discrete points of the initial trajectory are smoothed to reduce the robot motion jitter; combining with the industrial robot kinematic model and the rapidly-exploring random tree star algorithm to dynamically generate a local obstacle avoidance path, ensuring real-time avoidance of the interference areas of the fixture and the workpiece during the grinding process, and taking into account both processing efficiency and safety;

[0041] (4) Based on the pressure difference feedback of the force sensor, the trajectory depth is dynamically corrected to realize the closed-loop control of the grinding force, effectively coping with uneven burr thickness or surface undulation of the casting, and ensuring processing stability and finished product consistency. Description of the Drawings

[0042] Figure 1It is a step diagram of a trajectory generation method for deburring cast blanks. Detailed implementation manners

[0043] The following are specific embodiments of the present invention and in combination with the accompanying drawings, the technical solutions of the present invention will be further described, but the present invention is not limited to these embodiments.

[0044] To more intuitively reflect the actual application effect of the technical solution of the present invention, the following will describe the specific implementation manners in detail in combination with typical industrial scenarios. This embodiment takes the automatic deburring of a turbocharger housing blank as an example, and constructs a complete intelligent grinding system by integrating 3D vision perception, dynamic trajectory planning and real-time feedback control. In actual application, the system first obtains the three-dimensional data of the blank surface through a high-precision 3D scanning device, and accurately aligns the actual workpiece with the theoretical model by combining an improved point cloud registration method; then based on feature analysis and the user interaction interface, flexible segmentation and parameter configuration of the grinding area are realized, and customized trajectories are generated for different contour features (such as curved surface transition areas, edge burr areas); finally, through robot kinematics optimization and real-time force control feedback, the stability and consistency of the grinding process are ensured. The following will specifically describe around the hardware architecture, core algorithm implementation and multi-module cooperation mechanism, focusing on explaining the strategy for improving the accuracy of point cloud registration, the method for adaptively generating segmented trajectories, and the engineering application of dynamic obstacle avoidance and pressure compensation technology, so as to fully present the technical implementation path and innovative advantages of this solution in complex industrial scenarios. As Figure 1 shown, a trajectory generation method for deburring cast blanks proposed by the present invention mainly includes the steps:

[0045] S1: Scan the cast blank with a 3D line array camera to obtain the actual point cloud model, and extract the pre-entered standard point cloud model;

[0046] S2: Based on the improved iterative closest point algorithm, perform registration adjustment of the actual point cloud model and the standard point cloud model under coordinate system alignment;

[0047] S3: According to the FPFH feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model respectively, and search for key points of the standard point cloud model through KD tree matching, and divide the multi-contours of the actual point cloud model according to the distances between the key points;

[0048] S4: Receive the grinding parameters input by the user based on the results of each contour segmentation, and generate an initial trajectory according to the grinding parameters;

[0049] S5: Smooth the discrete trajectory points of the initial trajectory through B-spline interpolation, and generate a local obstacle avoidance path in combination with the industrial robot kinematics model and the rapidly-exploring random tree star algorithm;

[0050] S6: Output the final grinding trajectory to the control terminal and adjust the trajectory depth in real time according to the feedback of the force sensor.

[0051] In the scanning and data acquisition stage, the system completes the 3D data acquisition and preprocessing of the casting blank through an integrated hardware architecture. In specific implementation, the blank of the turbocharger housing is conveyed to the fixing tooling of the positioner by the feeding system, and the positioner adjusts the workpiece posture according to the preset program to ensure that the surface to be ground is vertically oriented towards the scanning direction. The 3D line array camera is mounted on the truss system and moves bidirectionally along the X-Y axis to scan the casting surface at a constant speed from multiple angles, covering key areas such as the rim, pores, and flange connection surface of the blank. During the scanning process, the line array camera captures point cloud data at a sampling rate of 2000 frames per second, synchronously triggers the annular LED light source to eliminate the interference of metal reflection, and performs real-time point cloud denoising to preliminarily filter out abnormal discrete points caused by dust or vibration. After the scanning is completed, the system starts the point cloud preprocessing process, including ground point cloud segmentation, non-grinding area shielding, and point cloud simplification.

[0052] Then, the system uses an improved Iterative Closest Point (ICP) algorithm to perform high-precision alignment between the actual point cloud model and the standard point cloud model. In specific implementation, first, the FPFH (Fast Point Feature Histograms) feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model are extracted respectively, and the initial paired points are screened through feature matching. The FPFH feature descriptor can effectively characterize the local geometric properties of the point cloud (such as curvature, normal direction), providing a reliable initial matching pair for subsequent registration. To improve the matching accuracy, the system introduces the Random Sample Consensus (RANSAC) algorithm to eliminate mis-matched points and ensure the uniformity of the spatial distribution of the paired points.

[0053] During the registration process, the system realizes coordinate system alignment by minimizing the error function of the rotation matrix and the translation vector . The specific formula is:

[0054]

[0055] where is the total number of data points of the point cloud, is the th three-dimensional point in the actual point cloud model, is the th three-dimensional point in the standard point cloud model. Through iterative optimization, the system gradually adjusts the rotation matrix and the translation vector , until the registration error is less than the preset error threshold. In each iteration, the system dynamically updates the matching point pairs and optimizes the registration result by combining local curvature constraints to avoid misalignment caused by similar local features.

[0056] To further improve the registration robustness, the system also introduces a multi-scale registration strategy: First, in the coarse registration stage, that is, in step S2, a low-resolution point cloud is used to quickly align the overall pose to eliminate mis-matched points. Subsequently, in the fine registration stage, that is, step S3, the high-resolution point cloud is switched to refine the local feature alignment. The finally output registration result will be displayed through the visualization interface, and the user can manually fine-tune the alignment accuracy of the key areas to ensure the reliability of the subsequent trajectory generation.

[0057] Among them, the high-resolution point cloud refining local feature alignment is realized by the KD tree (K-Dimensional Tree) search algorithm to accurately match key points and divide the contour segments. In the specific implementation, the FPFH feature library of the previously obtained actual point cloud model is compared with the pre-computed FPFH feature library of the standard point cloud model. By constructing the KD tree index structure of the standard point cloud model, the system quickly searches for the nearest neighbor key points corresponding to each point in the actual point cloud, and the matching formula is:

[0058]

[0059] where, in the formula, is the FPFH feature descriptor of the th point in the actual point cloud model, is the th three-dimensional point in the standard point cloud model, is the th point in the standard point cloud model, is the FPFH feature descriptor of the th point, is the index that finds the minimum Euclidean distance from all standard point cloud candidate points to , .

[0060] After the matching is completed, the system performs multi-contour segmentation based on the spatial distance between key points. For the set of continuously distributed matching key points in the actual point cloud, if the distance between adjacent key points exceeds the preset distance threshold, it is determined as the boundary point of the contour segmentation. At the same time, the system dynamically adjusts the segmentation radius, preferentially retaining the local continuity of high-curvature regions (such as edges and transition rounded corners) to avoid mis-segmentation caused by uneven point cloud density. The segmentation result is displayed through the visualization interface, and the user can manually adjust the segmentation boundary or merge adjacent regions to ensure the rationality of the grinding area division.

[0061] Preferably, to improve the reliability of segmentation, the system introduces a two-way verification mechanism: reverse search for the nearest neighbors of the actual point cloud key points in the standard point cloud model, and eliminate the point pairs with inconsistent two-way matching. Generally speaking, this stage combines automated segmentation with manual verification to achieve fine processing of complex contours and provide highly reliable input data for subsequent trajectory generation.

[0062] For the finally generated segmented contour schematic diagram, each paragraph is marked with a unique identifier, and during the parameter setting process, the user is supported to independently set the grinding depth for different paragraphs , the lateral expansion width and the quadratic function feed angle of the chamfering required area , where the coefficient is input by the user, is the final grinding trajectory, which is used to control the feed trajectory smoothness and cutting-in angle of the chamfering area, and avoid stress concentration or surface scratches caused by linear feeding.

[0063] Then the system dynamically adjusts the trajectory offset according to the parameters input by the user. For the reference trajectory (generated from the outer contour of the standard point cloud), the normal offset trajectory and the lateral expansion trajectory are calculated respectively by the following formulas:

[0064]

[0065] where, N is the normal direction of the point cloud, and T is the edge tangent direction. For the chamfering area, the system integrates the quadratic function into the trajectory interpolation algorithm to generate a continuously changing feed angle, ensuring the smooth movement of the grinding head in the curved surface transition area.

[0066] In addition, during the parameter input process, the system will also automatically detect the parameter rationality through the built-in verification mechanism: the grinding depth is restricted within the preset safety threshold (such as 0.1–2.0mm), and the quadratic function coefficient needs to meet the trajectory curvature continuity condition (such as non-abrupt second derivative). If conflicting parameters are detected (such as the lateral expansion width exceeding the fixture avoidance space), the system will prompt the user to correct in real time. After the configuration is completed, the initial trajectory is previewed in the form of a three-dimensional path, supporting the user to manually fine-tune the key point positions to generate parameterized initial trajectory data adapted to the multi-segment contour features, providing an input basis for subsequent optimization and collision avoidance planning.

[0067] Based on the obtained initial trajectory, the system smooths the initial trajectory through the B-spline interpolation algorithm and generates a local obstacle avoidance path by combining the kinematic constraints of the industrial robot and the Rapidly-exploring Random TreeStar (RRT*) algorithm. Traditional grinding trajectory planning methods usually rely on fixed path templates or simple linear interpolation, which are difficult to meet the smoothness requirements of complex contours and lack the effective ability to avoid dynamic interference. To address these problems, the present invention introduces the B-spline interpolation algorithm to optimize the initial trajectory. In specific implementation, the system reconstructs the discrete points of the initial trajectory by interpolating based on the B-spline basis function, and its mathematical expression is:

[0068]

[0069] Where is a parameter variable, is the optimized initial trajectory, is B-spline basis functions of order , are the control points dynamically adjusted according to the simulation results of the industrial robot kinematic model in the initial trajectory, is the total number of control points. The system optimizes the distribution of control points according to the robot kinematic model (such as joint acceleration and speed limits) to ensure that the interpolated trajectory meets the requirements of motion smoothness and reachability, and at the same time avoids trajectory mutations through curvature continuity detection. Compared with traditional linear interpolation, B-spline interpolation can significantly reduce trajectory jitter and improve the motion stability of the grinding head on complex curved surfaces.

[0070] Furthermore, to address the interference risk in complex fixture environments, the system constructs a geometric bounding box model of the robot, fixture, and workpiece, and verifies the feasibility of the trajectory pose based on the inverse kinematics solution. If potential interference is detected (such as collision between the grinding head and the fixture), the system triggers the RRT* (Rapidly-exploring Random TreeStar) algorithm to generate a local obstacle avoidance path. The RRT algorithm is a path planning method based on random sampling. By randomly generating nodes in the configuration space and constructing a tree-like path network, the path quality is gradually optimized. Compared with the traditional RRT algorithm, RRT* introduces a rewiring mechanism that can continuously optimize the path cost (such as path length and energy consumption) during the expansion process, thereby generating a more efficient obstacle avoidance path. In specific implementation, the system takes the minimum energy consumption as the optimization goal, screens the optimal detour path, and smoothly connects the obstacle avoidance path and the original trajectory through a cubic spline curve to ensure no jitter in the transition area.

[0071] Considering that traditional grinding systems often rely on the rigid execution of preset trajectories and are difficult to cope with the pressure fluctuations caused by uneven burr thickness or surface deformation of castings, which are prone to over-grinding or under-grinding defects. In response to this, after the optimized grinding trajectory is transmitted to the execution terminal through the industrial robot control interface, the system will analyze the trajectory data based on the robot kinematic model, convert it into servo motor instructions for each joint, and synchronously plan the attitude adjustment in the tool coordinate system (such as the inclination compensation of the grinding head). During the execution process, a high-frequency force sensor (sampling rate ≥ 1kHz) is mounted on the floating spindle at the end of the robot to monitor the normal contact pressure , and the data is fed back to the system. The system dynamically corrects the trajectory depth according to the pressure difference feedback by the force sensor, and its core formula is:

[0072]

[0073] Where, is the adjusted trajectory depth, is the original trajectory depth, is the proportionality coefficient (dynamically calibrated through an adaptive algorithm, such as gradient descent optimization based on historical pressure data), is the reference pressure (set according to material hardness and process requirements). When it is detected that continues to be lower than , the system determines that the burr thickness exceeds the expectation, and increases the grinding depth step by step according to gain, and vice versa to reduce the depth to avoid over-cutting. During the adjustment process, the depth correction amount is smoothed by a low-pass filter to prevent trajectory jitter caused by instantaneous pressure fluctuations. , and vice versa to reduce the depth to avoid over-cutting. During the adjustment process, the depth correction amount is smoothed by a low-pass filter to prevent trajectory jitter caused by instantaneous pressure fluctuations.

[0074] In the preferred embodiment, anomaly detection and fault tolerance processing are also introduced. If the pressure deviation continuously exceeds the threshold, an emergency pause is triggered, and a three-dimensional vision re-inspection module is started to re-scan the point cloud data of the abnormal area and update the local trajectory. At the same time, the system records the pressure-time curve and the depth correction log to provide data support for process optimization.

[0075] The present invention also includes a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method for generating a trajectory for deburring a casting blank are implemented.

[0076] It also includes a device for processing data, including:

[0077] A memory, on which a computer program is stored;

[0078] A processor, configured to execute the computer program in the memory to implement the steps of the method for generating a trajectory for deburring a casting blank.

[0079] In summary, a trajectory generation method for deburring cast blanks proposed by the present invention realizes the matching and segmentation of key points on the outer contour of the casting based on FPFH features and KD tree search. Combining the grinding parameters input by the user, it supports the generation of differentiated trajectories in different regions (such as chamfers and burr concentration areas), significantly improving the process adaptability to complex contours and the ability to respond to personalized requirements.

[0080] By combining the improved Iterative Closest Point algorithm (ICP) with the screening of FPFH feature descriptors and the elimination of mismatches by Random Sample Consensus (RANSAC), the registration accuracy between the actual point cloud and the standard model is significantly improved, effectively reducing the coordinate system alignment error caused by casting deformation, and avoiding over-grinding or equipment collision caused by model deviation.

[0081] The initial trajectory is smoothed by B-spline interpolation to reduce the robot motion jitter; combined with the industrial robot kinematic model and the Rapidly-exploring Random Tree Star algorithm to dynamically generate a local obstacle avoidance path, ensuring real-time avoidance of the interference areas of the fixture and the workpiece during the grinding process, taking into account both processing efficiency and safety. At the same time, the trajectory depth is dynamically corrected based on the pressure difference feedback of the force sensor to realize the closed-loop control of the grinding force, effectively coping with uneven burr thickness or surface undulations of the casting, and ensuring processing stability and product consistency.

[0082] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship and motion conditions between components in a specific posture (as shown in the attached drawings). If the specific posture changes, the directional indication will also change accordingly.

[0083] In addition, in the present invention, descriptions such as "first", "second", "one" are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0084] In the present invention, unless otherwise clearly defined and limited, terms such as "connection" and "fixation" should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two components or the interaction relationship between two components, unless otherwise clearly limited. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0085] In addition, the technical solutions between various embodiments of the present invention may be combined with each other, but it must be based on what can be achieved by those of ordinary skill in the art. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. A trajectory generation method for deburring a casting blank, characterized in that: Includes steps: S1: Scan the casting blank with a 3D linear array camera to obtain the actual point cloud model and extract the pre-recorded standard point cloud model; S2: Based on the improved iterative closest point algorithm, the actual point cloud model and the standard point cloud model are aligned under the coordinate system registration adjustment; S3: Based on the FPFH feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model, the key points matching the standard point cloud model are searched through the KD tree, and the multi-contour segmentation of the actual point cloud model is performed according to the distance between the key points; S4: receiving polishing parameters input by the user based on the segmentation results of each contour, and generating an initial trajectory according to the polishing parameters; S5: The initial trajectory is smoothed by B-spline interpolation, and the local obstacle avoidance path is generated by combining the industrial robot kinematic model and the fast expansion random tree star algorithm; S6: Output the final grinding track to the control terminal and adjust the track depth in real time according to the feedback from the force sensor; In the step S2, the improved iterative closest point algorithm is specifically: S21: extract the FPFH feature descriptors of the outer contours of the actual point cloud model and the standard point cloud model and select the initial pairing points; S22: The mismatched points are eliminated through the random sampling consistency algorithm, and the actual point cloud model and the standard point cloud model are aligned with the coordinate system through the following formula: In the formula, is the rotation matrix, is the translation vector, is the total amount of 3D points in the point cloud model, The actual point cloud model Three-dimensional points, The first 3D points; S23: Iterative optimization of rotation matrix and translation vectors , until the error of the registration adjustment is less than the preset error threshold.

2. A trajectory generation method for deburring a casting blank according to claim 1, characterized in that: In the step S3, the key points of the standard point cloud model searched by the KD tree are expressed as follows: In the formula, The actual point cloud model Points The FPFH feature descriptor, The first Three-dimensional points, The first Points The FPFH feature descriptor, To find the corresponding point in all the standard point cloud candidate points Index with minimum Euclidean distance .

3. A trajectory generation method for deburring a casting blank according to claim 1, characterized in that: In the step S4, the grinding parameters include grinding depth , Horizontal expansion width The quadratic function feed angle of the chamfering required area ,in, For user input coefficients, Polish the track for the final touch.

4. A trajectory generation method for deburring a casting blank as claimed in claim 3, characterized in that: The initial trajectory is expressed as the following formula: In the formula, is the normal offset trajectory, For horizontal expansion trajectory, is the reference trajectory obtained based on the standard point cloud model, N is the point cloud normal direction, and T is the edge tangent direction.

5. A trajectory generation method for deburring a casting blank according to claim 1, characterized in that: In the step S5, the smoothing of discrete trajectory points of the initial trajectory by B-spline interpolation is expressed as the following formula: In the formula, is a parameter variable, is the optimized initial trajectory, for The order is B-spline basis functions, is the control point in the initial trajectory that is dynamically adjusted according to the simulation results of the industrial robot kinematic model. is the total number of control points.

6. A trajectory generation method for deburring a casting blank according to claim 1, characterized in that: In the step S5, the generation of the local obstacle avoidance path by combining the industrial robot kinematics model and the RRT* algorithm is specifically as follows: Construct geometric bounding box models of robots, fixtures, and workpieces; Verify the feasibility of trajectory pose through inverse kinematics solution; If interference is detected, a fast expanding random tree star algorithm is used to generate a local obstacle avoidance path.

7. A trajectory generation method for deburring a casting blank according to claim 1, characterized in that: In the step S6, the real-time adjustment of the track depth according to the feedback of the force sensor is expressed as the following formula: In the formula, is the adjusted track depth, is the original track depth, is the proportionality coefficient, is the base pressure, is the real-time pressure of the force sensor.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of a trajectory generation method for deburring a casting blank as described in any one of claims 1 to 7 are implemented.

9. A device for processing data, characterized in that: include: a memory having a computer program stored thereon; A processor is used to execute the computer program in the memory to implement the steps of a trajectory generation method for deburring a casting blank as described in any one of claims 1 to 7.

Images