Visual guidance robot polishing path planning method and system

Abstract

The invention relates to the technical field of robot automatic polishing, in particular to a vision-guided robot polishing path planning method and system, and the method comprises the steps: obtaining a workpiece surface point cloud in real time through a three-dimensional vision sensor, generating a trajectory seed point with adaptive density in combination with curvature, a normal vector and a defect semantic tag, and carrying out the positioning of a workpiece surface point; and a polishing quality evaluation function is constructed based on force sense feedback and a visual reconstruction result, so that path online correction is realized. According to the method, the polishing path and the tool posture can be dynamically adjusted, high-precision differentiation processing is achieved on the free-form surface, the edge and the defect area, and the polishing consistency and efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and robotics technology, and specifically relates to a vision-guided robot polishing path planning method and system. Background Technology

[0002] Robotic grinding path planning technology is an important component of intelligent manufacturing and automated processing, widely used in industries such as automotive manufacturing, aerospace, and precision parts machining. Currently, it is generally implemented using preset trajectory programming or path generation methods based on offline models.

[0003] However, existing methods primarily rely on the workpiece's geometric model or a fixed teaching path. Their adaptability is limited when dealing with workpieces with complex surface topography, machining deviations, or clamping errors, potentially leading to uneven grinding or overcutting. Furthermore, actual grinding tasks typically involve free-form surfaces, edge features, and localized defect areas, while existing methods often rely on idealized models for path planning, making it difficult to fully integrate real-time surface information and easily resulting in omissions or redundant grinding of critical areas. In addition, existing systems have limited responsiveness in dynamically adjusting grinding parameters and paths, and lack a closed-loop feedback mechanism for grinding effects.

[0004] In summary, existing technologies still mainly rely on static models or manual teaching, failing to effectively combine visual perception with adaptive path optimization, resulting in limitations in polishing accuracy and automation levels. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a vision-guided robotic grinding path planning method and system. This invention utilizes a high-precision 3D vision sensor to collect real-time point cloud data of the workpiece surface, constructing a dynamically updated local geometric model to replace path generation methods relying on static offline models or manual teaching. It integrates surface curvature, normal vector distribution, and semantic labels of defect areas to generate a grinding trajectory seed point set with adaptive density and orientation characteristics. Based on real-time grinding force feedback and visual reconstruction results, it establishes grinding effect evaluation indicators and adjusts subsequent path parameters and tool posture online accordingly. A closed-loop control architecture tightly couples the perception, planning, and execution modules, enabling differentiated processing capabilities for the grinding path in free-form surfaces, edge transition zones, and local defect areas, avoiding omissions of key features or redundant processing.

[0006] A vision-guided robotic polishing path planning method includes the following steps: Before the grinding operation begins, the three-dimensional vision sensor is fixedly installed on the rigid bracket of the robot end effector or working unit so that its field of view covers the area to be processed. The three-dimensional vision sensor establishes a real-time data connection with the central controller through the industrial communication bus. The three-dimensional vision sensor is activated to scan the surface of the workpiece and acquire raw point cloud data. The central controller performs noise reduction, registration and downsampling preprocessing on the point cloud data to generate a structured surface point cloud. Based on the structured surface point cloud, the central controller calculates the local geometric features of each point, including principal curvature, Gaussian curvature, average curvature and surface normal vector, and identifies edge feature regions based on curvature change threshold. At the same time, it uses a deep learning model to perform semantic segmentation on the point cloud and marks local defect regions such as scratches, pits or oxide layers. The central controller generates non-uniformly distributed trajectory seed points on the workpiece surface based on the geometric features and semantic tags. The seed point density is increased in high curvature areas, edge transition zones, and defect marking areas, and decreased in flat areas. An initial grinding direction is assigned to each seed point, which is determined by the angle constraint between the local normal vector and the preset grinding tool axis. The central controller inputs the trajectory seed points into the path generation module based on spline interpolation to generate a continuously differentiable initial grinding path, and decomposes the path into several path segments, with each path segment associated with a corresponding local surface feature type. The robot control system receives the initial grinding path, drives the end effector to move along the path, and at the same time, the six-dimensional force / torque sensor installed between the end effector and the grinding tool collects the grinding contact force data in real time and transmits the data back to the central controller. During the polishing process, the central controller synchronously triggers the three-dimensional vision sensor to perform a secondary scan of the polished area, reconstruct the surface morphology after polishing, and perform differential analysis with the model before polishing to calculate the material removal distribution and surface flatness index. The central controller integrates the material removal distribution, surface flatness index and grinding contact force data from multiple sources to construct a grinding quality evaluation function. When the output value of the evaluation function exceeds the preset tolerance range, it is determined that the current path segment has over-cut or under-grind. For path segments that are overcut or under-grinded, the central controller calls the local replanning module to regenerate the optimized sub-path based on the original seed point, combined with the current surface condition and tool wear compensation parameters, and inserts the sub-path into the corresponding position in the original path sequence. The central controller sends the updated complete polishing path to the robot motion controller through the real-time communication interface. The latter uses a time-optimal trajectory planning algorithm to generate joint space instructions, driving the robot to execute the corrected polishing action. Throughout the entire polishing cycle, the central controller continuously performs surface scanning, quality assessment, and path correction operations in a loop until the polishing quality assessment function of all areas falls within the preset tolerance range, at which point the operation terminates.

[0007] A vision-guided robotic polishing path planning system includes a 3D vision sensor, a six-dimensional force / torque sensor, a central controller, a robot motion controller, and an end effector. The 3D vision sensor is fixedly mounted on the flange side or an independent rigid bracket of the robot end effector, with its optical axis forming a preset angle with the axis of the polishing tool to ensure unobstructed observation of the processing area during polishing. The six-dimensional force / torque sensor is connected in series between the end effector and the polishing tool, and its output is connected to the signal acquisition module of the central controller via an analog or digital interface. The central controller integrates a point cloud processing unit, a geometric feature extraction module, a semantic segmentation neural network, a path generation engine, a quality assessment module, and a local replanning unit. These modules interact with each other through shared memory or a message queue. The robot motion controller is connected to the central controller via an industrial Ethernet network, receiving path commands and outputting servo drive signals to the robot's joint motors. The end effector is equipped with an active compliance mechanism, whose stiffness parameters are dynamically set by the central controller based on the characteristics of the current polishing area.

[0008] Compared with the prior art, the present invention has the following beneficial effects: This invention abandons the traditional method of relying on idealized CAD models or fixed teaching paths. It constructs a fusion representation of the workpiece surface geometry and semantic information in real time through a three-dimensional vision sensor, so that path planning is directly based on the real physical state, effectively overcoming the problem of insufficient adaptability caused by clamping errors, processing deviations or surface morphology complexity. This invention introduces a curvature-driven seed point density control mechanism and a defect area priority coverage strategy in the path generation stage to ensure higher trajectory resolution and targeted polishing strategies in the free surface transition area, sharp edges and local defects, avoiding the waste of efficiency caused by omission of key areas or global uniform polishing. This invention constructs a multimodal grinding quality assessment system that includes visual reconstruction, force feedback and material removal model, and realizes online local replanning of the path on this basis, forming a closed-loop control loop of perception-decision-execution, which significantly improves the system's response to dynamic disturbances and uncertainties. This invention couples and constrains the posture of the grinding tool with the local surface normal vector, and combines the stiffness adjustment of the active compliance mechanism to keep the grinding contact force stable in different areas, preventing overcutting or tool jump caused by abrupt changes in the normal vector, and ensuring surface consistency. The system architecture of this invention adopts a modular design, and each functional unit is interconnected through standardized interfaces, which facilitates integration into different brands of industrial robot platforms. It has good versatility and engineering feasibility, and is suitable for a variety of high-precision automated grinding scenarios such as automotive body weld grinding, aero-engine blade polishing, and mold repair. Attached Figure Description

[0009] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0010] Figure 1 This is a schematic diagram of a vision-guided robot polishing path planning method according to one embodiment. Detailed Implementation

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0012] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, 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 invention pertains.

[0013] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0014] Where there is no conflict, the embodiments and features described in this application may be combined with each other. Example 1

[0015] A vision-guided robotic polishing path planning method, such as Figure 1 As shown, it includes the following steps: Before the grinding operation begins, the three-dimensional vision sensor is fixedly installed on the side of the flange of the robot end effector or on a rigid bracket in the working unit so that its field of view completely covers the area to be processed. The three-dimensional vision sensor establishes a real-time data communication link with the central controller through industrial Ethernet or EtherCAT bus. The three-dimensional vision sensor is activated to perform non-contact scanning of the workpiece surface to obtain raw point cloud data. The central controller then performs statistical outlier removal, ICP registration, and voxel grid downsampling preprocessing on the point cloud data to generate a structured surface point cloud. Based on the structured surface point cloud, the central controller calls the geometric feature extraction module to calculate the principal curvature of each sampling point. and Gaussian curvature Mean curvature and the unit surface normal vector n, and set the curvature abrupt change threshold. When adjacent points When the point cloud is identified as an edge feature region, the central controller loads the pre-trained PointNet++ semantic segmentation network to perform pixel-level classification of the point cloud and output semantic labels for three types of defect regions: scratches, pits, and oxide layers. The central controller generates a non-uniformly distributed set of trajectory seed points on the workpiece surface based on the geometric features and semantic tags, wherein... In high-curvature regions, edge transition zones, and regions semantically labeled as defects, the seed point density is set to 16 points per square centimeter. The flat area is reduced to 4 points per square centimeter, and an initial grinding direction vector di is assigned to each seed point. This vector is determined by the angle constraint between the local normal vector ni and the preset grinding tool axis atool, satisfying the following conditions: ; The central controller inputs the trajectory seed points into the path generation engine based on cubic B-spline interpolation to generate an initial polishing path that is continuously differentiable and has continuous curvature. The path is then divided into several path segments according to the surface feature type, and each path segment is associated with a feature identifier, including "high curvature", "edge", "defect" or "flat". The robot motion controller receives the spatial coordinate sequence and attitude commands of the initial grinding path, driving the six-axis robotic arm to move along the path. Simultaneously, a six-dimensional force / torque sensor installed between the end flange and the grinding tool base collects contact force data in real time at a sampling rate of 1 kHz. It is transmitted back to the central controller via analog differential signals or the CANopen protocol; During the polishing process, after completing each path segment, the central controller synchronously triggers the 3D vision sensor to perform a secondary scan of the polished area, reconstructs the point cloud of the polished surface, and performs differential analysis after ICP alignment with the corresponding point cloud of the area before polishing to calculate the material removal depth distribution. Surface flatness index ; The central controller removes the material at different depths. Surface roughness Contact force with normal direction Perform weighted fusion to construct a polishing quality evaluation function. ,in N is the reference contact force, when When this happens, it is determined that the current path segment has either been overcut or undercut. For path segments with overcutting or under-wearing, the central controller invokes a local replanning unit to overlay the current surface state point cloud onto the original seed point, and incorporates tool wear compensation parameters. The optimized sub-path is regenerated, which uses a higher density of seed points (density increased by 50%) and the grinding direction is adjusted to compensate for insufficient material removal. Then the sub-path is inserted into the corresponding position in the original path sequence. The central controller sends the updated complete polishing path to the robot motion controller via a real-time UDP interface. The latter uses a time-optimal trajectory planning algorithm (such as TOPP-RA) to generate a sequence of instructions with limited joint space velocity and acceleration, driving the servo motors to execute the corrected polishing action. Throughout the entire polishing cycle, the central controller continuously performs surface scanning, quality assessment, and path correction operations in a loop until the polishing quality assessment function Q for all path segments is satisfied. The operation is terminated.

[0016] In this embodiment, the PointNet++ semantic segmentation network is pre-trained on a dataset containing 10,000 sets of labeled point clouds. The dataset covers typical defect samples of aerospace aluminum alloy blades, automotive welds, and mold steel surfaces, enabling it to accurately identify sub-millimeter scratches and localized oxidation areas.

[0017] Specifically, for free-form surface workpieces (such as turbine blades) and planar welded workpieces (such as car body seams), the system automatically switches between different curvature thresholds and seed point density strategies: for blade-type workpieces, the high curvature threshold is set to... The seed point density in the defect area is increased to 25 points / cm²; for welded workpieces, edge recognition prioritizes the use of the abrupt change criterion of the included angle of the normal vector. Furthermore, a minimum seed point spacing of 1.2 mm is forcibly set in the weld toe area.

[0018] The 3D vision sensor can be a laser line scanner or a structured light camera, and its installation position must ensure that the optical axis is perpendicular to the axis of the grinding tool. Angle is used to avoid sanding dust obstructing the tool and interfering with its operation.

[0019] The compensation for workpiece clamping errors is achieved through an initial global scan: before formal grinding, the system performs a low-resolution scan of the entire workpiece, performs coarse registration with the theoretical CAD model, and calculates the rigid body transformation matrix. This transformation is then applied to all subsequent local path generation, thereby eliminating path offset caused by fixture positioning deviation.

[0020] For example, a certain aero-engine blade workpiece is deflected by 2° due to clamping. The system detects this deflection during the initial scan and multiplies the coordinates of all seed points by the left during the path generation stage. This ensures that the polishing path strictly conforms to the actual physical surface.

[0021] If the workpiece is a weld seam of a car B-pillar, and there are multiple microcracks and spatter residues on its surface, the semantic segmentation network marks it as a "defect" area. The system automatically encrypts the trajectory in these areas and extends the dwell time to ensure that surface defects are completely removed.

[0022] In this embodiment, the process of constructing the polishing quality evaluation function specifically includes: First, based on the standard deviation of the material removal depth in differential point cloud computing. This reflects the uniformity of removal; if This indicates that there is local over-cutting or under-grinding.

[0023] Secondly, the surface roughness is calculated using the moving window method. The window size is 5 mm × 5 mm. If the surface finish is not up to standard, then it is determined that the surface smoothness is not up to standard.

[0024] Finally, monitor the normal contact force. With set value The absolute value of the deviation, if If the contact lasts for more than 0.5 seconds, it is considered an abnormal contact condition.

[0025] This embodiment integrates geometric, visual, and force information to construct quantifiable grinding quality indicators, overcoming the limitations of traditional methods that rely on manual visual inspection or offline measurement due to their lag. This provides a reliable basis for online closed-loop control and significantly improves the consistency and reliability of grinding complex curved surfaces.

[0026] In this embodiment, the process of generating and assigning trajectory seed points includes: By using a curvature adaptive sampling algorithm, the number of sampling points is dynamically increased in regions where the absolute value of Gaussian curvature is greater than a threshold, ensuring that the surface details are fully represented. Using a semantic segmentation result mask, additional seed points are forcibly inserted within the defect label coverage area, with the point spacing not exceeding 60% of the tool's effective radius. Calculate a local coordinate system for each seed point, with its Z-axis aligned to the surface normal vector. The X-axis is along the direction of maximum principal curvature. The Y-axis is determined by the cross product. Determine the grinding direction. In this local coordinate system, it is set to be tilted at a 15° angle to the Z-axis and offset along the positive Y-axis to form a cutting flow direction that is conducive to chip removal.

[0027] Next, the path generation engine dynamically adjusts the spline interpolation parameters according to task requirements: if the current path segment is associated with the "defect" label, the interpolation tension coefficient is reduced to 0.2 to improve path compliance and adapt to local morphology; if it is associated with the "edge" label, the tension coefficient is increased to 0.8 to suppress path oscillation and ensure the accuracy of sharp edge contours.

[0028] In this embodiment, the process of the local replanning module generating optimized sub-paths includes: jointly optimizing the current surface reconstructed point cloud with the original seed point set, in areas with insufficient material removal. Insert new seed points and adjust the grinding force command based on the stiffness feedback of the active compliance mechanism to form a closed-loop verification chain of "perception-evaluation-correction," accurately locating the specific spatial position of over-cutting or under-cutting, and generating diagnostic labels containing the problem type (such as "under-cutting," "over-cutting," or "jumping"). For example, if a certain area... and If the target is 0.12 mm, it is judged as "under-grinding", and the system automatically generates a high-density, low-speed, and force-enhancing supplementary grinding sub-path.

[0029] This embodiment can adaptively adjust the path generation strategy and polishing parameters according to the surface condition and task type, so that high-value areas are given priority. Through multimodal perception and closed-loop replanning mechanism, the polishing path is dynamically aligned with the real physical surface, which can accurately locate specific quality problems at the millimeter level and generate structured diagnostic reports, providing data support for process traceability and parameter optimization.

[0030] The above-mentioned solution provides an automated grinding solution with high adaptability and strong robustness for the field of intelligent manufacturing through a collaborative mechanism of geometric-semantic fusion modeling, multi-source quality assessment and online local replanning. It significantly promotes the paradigm shift of robotic grinding from pre-programmed open-loop operation to perception-driven closed-loop control.

[0031] The process of determining whether the polishing path needs to be modified and the modification strategy includes: based on the output value and component contributions of the polishing quality evaluation function Q, if... And mainly composed of If the problem is caused by exceeding the limit, it is determined to be due to insufficient path density, and the seed points need to be densified; if it is mainly caused by... If the problem is caused by exceeding the limit, it is determined to be due to tool wear or excessive feed rate, requiring speed reduction and tool radius compensation; if it is mainly caused by... If the problem is caused by exceeding the limit, it is determined to be a mismatch in compliance stiffness, and the stiffness parameters of the active compliance mechanism need to be adjusted.

[0032] In this embodiment, based on a distributed message queue architecture, each path correction event is encapsulated as a structured log record, the content of which includes: Metadata: Path segment ID, correction timestamp, trigger reason code; Incremental content: Only the corrected sub-path coordinate sequence (not the full path) is stored, reducing the data volume by more than 70% compared to the original path.

[0033] In this embodiment, a visual monitoring interface is generated based on the polishing process data. For example, a heat map can be used to display the material removal depth distribution, highlighted in red. The system can identify areas and label problem types, and can also recommend process parameters based on historical data, such as when tool wear is detected as a cause of problem. If the quality continues to rise, it is recommended to replace the grinding wheel in advance or increase the coolant flow rate. The system can also trace the historical quality evolution trend of any grinding area through a time-series database. Example 2

[0034] A vision-guided robotic grinding path planning system includes a 3D vision sensor, a six-dimensional force / torque sensor, a central controller, a robot motion controller, and an end effector. The 3D vision sensor is rigidly fixed to the side wall of the robot's sixth-axis flange using a mechanical clamp, with its optical axis forming a 45° angle with the grinding tool's rotation axis to ensure unobstructed vision within the tool's working envelope. The six-dimensional force / torque sensor is connected in series between the end effector's output flange and the grinding tool's mounting base. Its strain gauge bridge output is converted into a ±10V analog signal by a signal conditioning circuit and connected to the central controller's high-speed ADC acquisition card. The central controller integrates a multi-core CPU and a GPU coprocessor, and internally deploys a point cloud processing unit, a geometric feature extraction module, a PointNet++ semantic segmentation neural network, a B-spline path generation engine, a multi-source quality assessment module, and a local replanning unit. All software modules utilize ROS... The two nodes interact with each other using the DDS protocol for low-latency data exchange. The robot motion controller is connected to the central controller via a gigabit industrial Ethernet, receiving Cartesian space path points and tool posture quaternions, and generating position, velocity, and acceleration commands for each joint based on inverse kinematics calculations. These commands are amplified by servo drivers and output to the six-axis servo motors. The end effector incorporates an electro-pneumatic compliance mechanism with an equivalent stiffness of [missing information]. The air pressure is regulated by a proportional valve, and the central controller dynamically sets the pressure based on the characteristics of the current path segment. Value: Set to "High Curvature" in the region Set in the "flat" area . Example 3

[0035] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps in the method mentioned in Embodiment 1.

[0036] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0037] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0038] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0039] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0040] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0041] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principles of this invention are further supplemented below with a specific application scenario.

[0042] Taking the automated grinding task of titanium alloy turbine blades for aero-engines as an example, this workpiece has a highly complex free-form surface, sharp leading and trailing edges, as well as local pits and oxide layers left over from the casting process. Traditional teaching or offline programming methods are difficult to simultaneously achieve edge conformation preservation and defect removal. In the implementation of this invention, the structured light 3D vision sensor 1 is first fixed to the side wall of the sixth axis flange of the robot using a rigid clamp. Its optical axis forms a 45° angle with the axis of the rotating grinding wheel tool mounted at the end, ensuring no self-occlusion within the working envelope of the grinding wheel and complete coverage of the blade's leading to trailing edge region. Subsequently, the central controller 2 activates sensor 1 to perform an initial global scan of the blade surface, acquiring a low-resolution point cloud, and performing ICP coarse registration with the blade's theoretical CAD model to calculate the overall pose offset matrix caused by clamping. This matrix is ​​used to correct all subsequent local path coordinate systems, thereby eliminating the impact of fixture positioning errors on path accuracy.

[0043] Entering the fine scanning stage, the central controller 2 controls the robot to slowly move sensor 1 along the normal direction of the blade surface, collecting high-density raw point cloud data. After statistical outlier removal and voxel downsampling, the point cloud is input into the geometric feature extraction module, which calculates the Gaussian curvature K and the unit normal vector n point by point. When adjacent points... At that moment, the system determines whether the region is the leading or trailing edge; simultaneously, the pre-trained PointNet++ network performs semantic reasoning on the point cloud, classifying the surface into three categories: "normal," "pitted," and "oxide layer." Based on this fused information, the central controller 2... High-density trajectory seed points are generated for all semantically labeled "dimples," with a density of 25 points / cm². Meanwhile, in the flat area of ​​the middle of the leaf blade... Only 4 points / cm² are used. The grinding direction vector for each seed point. In its local coordinate system, set the Z-axis alignment. The X-axis is along the direction of the maximum principal curvature, and the Y-axis is determined by the cross product. The grinding direction is at a 15° angle to the Z-axis and offset along the positive Y-axis to guide the grinding debris out along the curvature flow line and prevent accumulation that could cause secondary scratches.

[0044] The initial grinding path is generated by cubic B-spline interpolation, and path segments are labeled as "high curvature" or "defect" according to their feature type. After receiving the path command, the robot motion controller 3 drives the six-axis robotic arm to perform the grinding action. At this time, the six-dimensional force / torque sensor 4, installed between the end flange and the grinding wheel base, collects the contact force at a frequency of 1 kHz and transmits the normal component back in real time. After completing a section of the path, the central controller 2 immediately triggers sensor 1 to perform a secondary scan of the polished area, reconstruct the point cloud after polishing, and perform differential calculations after ICP fine registration with the corresponding point cloud before polishing to obtain the material removal depth distribution. If a certain area (Target value 0.12 mm) and surface roughness Then the quality assessment function If the output value exceeds the threshold of 0.25, the system determines it as "under-wearing".

[0045] For this region, the local replanning unit is based on the original seed point, overlaid with the current surface point cloud, and incorporates tool wear compensation. ,exist New seed points are inserted into the sub-region, increasing the density by 50%. Simultaneously, the grinding speed is reduced by 20%, the contact force command is increased to 35 N, and the grinding direction is adjusted to be perpendicular to the gradient direction where material removal is insufficient. The generated optimized sub-path is inserted into the corresponding position of the original path sequence and sent to the robot motion controller 3 via the UDP interface. The latter uses the TOPP-RA algorithm to replan the time-optimal joint trajectory and executes corrective actions while ensuring acceleration constraints.

[0046] Throughout the entire operation cycle, the aforementioned "scan-evaluate-replan" loop continues to run until all path segments are... Ultimately, the system not only achieved a blade leading edge profile error of less than ±0.05 mm, but also improved surface roughness. In addition to meeting process requirements, the invention also records each correction event through distributed logging, including path segment ID, trigger reason code, and incremental sub-path coordinates, providing a traceable data foundation for subsequent process optimization. This process fully demonstrates the technical essence of this invention: real-time alignment of the grinding path with the actual physical surface through geometric-semantic fusion modeling, multi-source closed-loop feedback, and local dynamic replanning. This effectively solves the problem of high-precision automated grinding of complex free-form surface workpieces under conditions of clamping deviation and local defects.

Claims

1. A vision-guided robotic polishing path planning method, characterized in that, Includes the following steps: Before the grinding operation begins, the three-dimensional vision sensor is fixedly installed on the rigid bracket of the robot end effector or working unit so that its field of view covers the area to be processed. The three-dimensional vision sensor establishes a real-time data connection with the central controller through the industrial communication bus. The three-dimensional vision sensor is activated to scan the surface of the workpiece and acquire raw point cloud data. The central controller performs noise reduction, registration and downsampling preprocessing on the point cloud data to generate a structured surface point cloud. Based on the structured surface point cloud, the central controller calculates the local geometric features of each point, including principal curvature, Gaussian curvature, average curvature and surface normal vector, and identifies edge feature regions based on curvature change threshold. At the same time, it uses a deep learning model to perform semantic segmentation on the point cloud and marks local defect regions such as scratches, pits or oxide layers. The central controller generates non-uniformly distributed trajectory seed points on the workpiece surface based on the geometric features and semantic tags. The seed point density is increased in high curvature areas, edge transition zones, and defect marking areas, and decreased in flat areas. An initial grinding direction is assigned to each seed point, which is determined by the angle constraint between the local normal vector and the preset grinding tool axis. The central controller inputs the trajectory seed points into the path generation module based on spline interpolation to generate a continuously differentiable initial grinding path, and decomposes the path into several path segments, with each path segment associated with a corresponding local surface feature type. The robot control system receives the initial grinding path, drives the end effector to move along the path, and at the same time, the six-dimensional force / torque sensor installed between the end effector and the grinding tool collects the grinding contact force data in real time and transmits the data back to the central controller. During the polishing process, the central controller synchronously triggers the three-dimensional vision sensor to perform a secondary scan of the polished area, reconstruct the surface morphology after polishing, and perform differential analysis with the model before polishing to calculate the material removal distribution and surface flatness index. The central controller integrates the material removal distribution, surface flatness index and grinding contact force data from multiple sources to construct a grinding quality evaluation function. When the output value of the evaluation function exceeds the preset tolerance range, it is determined that the current path segment has over-cut or under-grind. For path segments that are overcut or under-grinded, the central controller calls the local replanning module to regenerate the optimized sub-path based on the original seed point, combined with the current surface condition and tool wear compensation parameters, and inserts the sub-path into the corresponding position in the original path sequence. The central controller sends the updated complete polishing path to the robot motion controller through the real-time communication interface. The latter uses a time-optimal trajectory planning algorithm to generate joint space instructions, driving the robot to execute the corrected polishing action. Throughout the entire polishing cycle, the central controller continuously performs surface scanning, quality assessment, and path correction operations in a loop until the polishing quality assessment function of all areas falls within the preset tolerance range, at which point the operation terminates.

2. The vision-guided robot polishing path planning method as described in claim 1, characterized in that, The three-dimensional vision sensor is fixedly installed on the side of the flange of the robot end effector or on an independent rigid bracket. Its optical axis is at a preset angle with the axis of the grinding tool to ensure unobstructed observation of the processing area during the grinding process.

3. The vision-guided robot polishing path planning method as described in claim 1, characterized in that, The deep learning model is a semantic segmentation neural network based on point clouds, and its training dataset contains point cloud samples of workpiece surfaces labeled with scratches, pits, or oxide layers.

4. The vision-guided robot polishing path planning method as described in claim 1, characterized in that, The density of the trajectory seed points and the local Gaussian curvature and mean curvature The weighted combination is directly proportional, specifically expressed as ,in and These are preset weighting coefficients.

5. The vision-guided robot polishing path planning method as described in claim 1, characterized in that, The initial polishing direction is determined by the local surface normal vector n and the axis of the polishing tool. The angle between Constraints, satisfaction ,in To preset the target angle, This is the allowable deviation threshold.

6. The vision-guided robot polishing path planning method as described in claim 1, characterized in that, The polishing quality evaluation function Q is defined as follows: ; Where R represents the material removal rate distribution. For normal grinding force, The reference normal force is S, and the surface flatness index is S. These are non-negative weighting coefficients.

7. The vision-guided robot polishing path planning method as described in claim 1, characterized in that, When generating the optimized sub-path, the local replanning module introduces tool wear compensation parameters. It is used to adjust the effective radius of the grinding tool.

8. A vision-guided robotic polishing path planning system, characterized in that, The system includes a 3D vision sensor, a six-dimensional force / torque sensor, a central controller, a robot motion controller, and an end effector. The 3D vision sensor is fixedly mounted on the flange side or an independent rigid bracket of the robot end effector, with its optical axis forming a preset angle with the axis of the grinding tool. The six-dimensional force / torque sensor is connected in series between the end effector and the grinding tool, and its output is connected to the signal acquisition module of the central controller via an analog or digital interface. The central controller incorporates a point cloud processing unit, a geometric feature extraction module, a semantic segmentation neural network, a path generation engine, a quality assessment module, and a local replanning unit. The robot motion controller is connected to the central controller via an industrial Ethernet network. The end effector is equipped with an active compliance mechanism.

9. A vision-guided robotic polishing path planning system as described in claim 8, characterized in that, The stiffness parameters of the active compliance mechanism are dynamically set by the central controller based on the local geometric characteristics of the current grinding area.

10. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps of the method according to any one of claims 1 to 7.

Images