Multi-stage polishing method and system for continuous surface in automobile bumper

Abstract

The application discloses a kind of multistage polishing methods and systems of continuity surface in automobile bumper, and the application relates to the technical field of surface polishing, and the corresponding initial tool path is determined according to the identification of the continuity surface of automobile bumper, and the corresponding polishing track is determined in combination with the parameter combination of polishing tool;Corresponding polishing tool and process parameter combination are distributed for rough grinding section, semi-fine grinding section and fine grinding section respectively.Mark the deviation content between actual contact force and target contact force, and dynamically adjust the normal feed depth and attitude angle of polishing head along the deviation content;Vibration signal and temperature signal are collected in real time during polishing process, and polishing quality evaluation matrix is constructed in combination with image data of machined area;If the waviness index in polishing quality evaluation matrix exceeds the preset index threshold, then trigger online compensation mechanism, and automatically correct the trajectory offset amount of subsequent unprocessed area and adjust contact pressure parameter, improve the accuracy of multistage polishing of continuity surface.

Description

Technical Field

[0001] This invention relates to the technical field of curved surface polishing, and more particularly to a multi-stage polishing method and system for continuous curved surfaces in automobile bumpers. Background Technology

[0002] With the rapid development of the automotive industry, consumers have increasingly stringent requirements for the quality of car exteriors. As one of the main external body panels, the surface finish of the car bumper directly determines the overall visual appeal and market competitiveness of the vehicle. However, most existing car bumpers are made of modified plastics such as TPO. During injection molding, these materials are affected by uneven cooling, mold precision, and differences in shrinkage, inevitably resulting in defects such as weld lines, shrinkage marks, or parting lines. Therefore, high-precision automated grinding is essential before painting the car bumper surface.

[0003] For grinding operations on car bumpers, industrial sites mainly use teach-in programming or offline programming to control industrial robots to execute preset trajectories. Car bumpers usually have complex shapes and continuous curved surface features with large curvature changes. Traditional path planning methods often cannot accurately fit the geometric features of these complex curved surfaces. In addition, existing grinding processes often use a single or simple grading method, lacking a collaborative design of trajectory and parameters for the entire process of rough grinding, semi-fine grinding and fine grinding, which cannot guarantee the accuracy of multi-level grinding of continuous curved surfaces. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a multi-level grinding method and system for continuous curved surfaces in automobile bumpers.

[0005] This invention provides a multi-stage polishing method for continuous curved surfaces in automobile bumpers, including:

[0006] The machined area of ​​the car bumper is collected, and a continuous surface is determined based on the identification of the machined area; the corresponding initial tool path is determined according to the identification of the continuous surface, and the corresponding grinding trajectory is determined by combining the parameter combination of the grinding tool. The polishing path is divided into coarse polishing, semi-fine polishing and fine polishing, and corresponding polishing tools and process parameter combinations are assigned to the coarse polishing, semi-fine polishing and fine polishing sections respectively. The process tolerance of the later polishing section covers the surface texture depth left by the previous polishing section. The end effector of the grinding robot moves along the grinding trajectory according to the combination of process parameters, and monitors the actual contact force between the grinding tool and the bumper surface in real time during the grinding process. It marks the deviation between the actual contact force and the target contact force, and dynamically adjusts the normal feed depth and attitude angle of the grinding head along the deviation. Vibration and temperature signals are collected in real time during the polishing process, and a polishing quality evaluation matrix is ​​constructed by combining the image data of the processed area. If the waviness index corresponding to the polishing quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-level polishing of the entire continuous surface is completed.

[0007] This invention provides a multi-stage polishing system for continuous curved surfaces in automobile bumpers, which is applied to the aforementioned multi-stage polishing method for continuous curved surfaces in automobile bumpers; the multi-stage polishing system for continuous curved surfaces in automobile bumpers includes: The grinding trajectory module is used to acquire the area to be processed of the car bumper, determine the continuous surface based on the identification of the area to be processed, determine the corresponding initial tool path based on the identification of the continuous surface, and determine the corresponding grinding trajectory by combining the parameter combination of the grinding tool. The process parameter module is used to divide the grinding trajectory into rough grinding, semi-fine grinding and fine grinding sections, and assign corresponding grinding tools and process parameter combinations to the rough grinding, semi-fine grinding and fine grinding sections respectively. The process tolerance of the later grinding section covers the surface texture depth left by the previous grinding section. The deviation content module is used for the end effector of the grinding robot to move along the grinding trajectory according to the combination of process parameters, and to monitor the actual contact force between the grinding tool and the surface of the bumper in real time during the grinding process, mark the deviation content between the actual contact force and the target contact force, and dynamically adjust the normal feed depth and attitude angle of the grinding head along the deviation content. The multi-stage grinding module is used to collect vibration and temperature signals in real time during the grinding process, and to construct a grinding quality evaluation matrix by combining the image data of the processed area. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-stage grinding of the entire continuous surface is completed.

[0008] Compared with the prior art, the beneficial effects of the present invention are: (1) Collect the processing area of ​​the car bumper and determine the continuous surface based on the identification of the processing area; determine the corresponding initial tool path according to the identification of the continuous surface, and determine the corresponding grinding trajectory in combination with the parameter combination of the grinding tool; divide the grinding trajectory into rough grinding section, semi-fine grinding section and fine grinding section, and assign corresponding grinding tools and process parameter combinations to the rough grinding section, semi-fine grinding section and fine grinding section respectively, and the process tolerance of the later grinding section covers the surface texture depth left by the previous grinding section, thus introducing the grinding trajectory and further controlling the grinding tools and process parameter combinations.

[0009] (2) The end effector of the grinding robot moves along the grinding trajectory according to the combination of process parameters, and monitors the actual contact force between the grinding tool and the bumper surface in real time during the grinding process. It marks the deviation between the actual contact force and the target contact force, and dynamically adjusts the normal feed depth and attitude angle of the grinding head along the deviation. Vibration signals and temperature signals are collected in real time during the grinding process, and a grinding quality evaluation matrix is ​​constructed by combining the image data of the processed area. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, the online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-level grinding of the entire continuous surface is completed. The deviation is further controlled, and the grinding quality evaluation matrix and the online compensation mechanism are fully considered. The multi-level grinding of the entire continuous surface is realized, and the accuracy of the multi-level grinding of the continuous surface is improved. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the multi-stage grinding method for continuous curved surfaces in a car bumper according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 of the multi-level polishing method for continuous curved surfaces in a car bumper according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S12 of the multi-level polishing method for continuous curved surfaces in a car bumper according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating step S13 of the multi-level polishing method for continuous curved surfaces in a car bumper according to an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 of the multi-level polishing method for continuous curved surfaces in a car bumper according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the structure of a multi-stage polishing system for continuous curved surfaces in a car bumper according to an embodiment of the present invention. Detailed Implementation

[0011] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0012] Please see Figures 1 to 6 A multi-level grinding method for continuous curved surfaces in car bumpers, applied to curved surface grinding scenarios; the multi-level grinding method for continuous curved surfaces in car bumpers includes: Step S11: Collect the area to be processed of the car bumper, and determine the continuous surface based on the identification of the area to be processed; determine the corresponding initial tool path according to the identification of the continuous surface, and determine the corresponding grinding trajectory by combining the parameter combination of the grinding tool. Step S12: Divide the grinding path into a rough grinding section, a semi-fine grinding section and a fine grinding section, and assign corresponding grinding tools and process parameter combinations to the rough grinding section, the semi-fine grinding section and the fine grinding section respectively, and the process tolerance of the later grinding section covers the surface texture depth left by the previous grinding section. Step S13: The end effector of the grinding robot moves along the grinding trajectory according to the combination of process parameters, and monitors the actual contact force between the grinding tool and the surface of the bumper in real time during the grinding process, marks the deviation between the actual contact force and the target contact force, and dynamically adjusts the normal feed depth and attitude angle of the grinding head along the deviation. Step S14: During the grinding process, vibration and temperature signals are collected in real time, and a grinding quality evaluation matrix is ​​constructed by combining the image data of the processed area. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-level grinding of the entire continuous surface is completed.

[0013] refer to Figure 2 In step S11, the specific steps are as follows: S111: The surface of the car bumper is dynamically scanned using a line laser scanner to collect three-dimensional point cloud data. Based on the three-dimensional point cloud data, a three-dimensional mesh model of the area to be processed is reconstructed. The principal curvature and Gaussian curvature of each point on the three-dimensional mesh model are marked. The corresponding continuous surface is identified based on the continuity of the rate of change of curvature. S112: Based on the feature recognition of the continuous surface, the corresponding geometric feature vector is determined, and the corresponding initial tool path is generated by combining the adaptive path planning mechanism. The initial tool path uses B-spline curves to smooth the transition in the curvature change region of the continuous surface to avoid motion impact caused by path inflection points. S113: Based on the geometric parameters of the grinding tool, material properties, and expected material removal rate, determine the contact interference between the grinding tool envelope and the surface to be machined, and then determine the final grinding trajectory, which includes the tool axis vector, step distance, and line distance.

[0014] In the embodiments of this application, a line laser scanner is used to dynamically scan the surface of a car bumper to collect three-dimensional point cloud data. A three-dimensional mesh model of the area to be processed is reconstructed based on the three-dimensional point cloud data. The principal curvature and Gaussian curvature of each point on the three-dimensional mesh model are marked. The corresponding continuous surface is identified based on the continuity of the rate of change of curvature, thus introducing a continuous surface.

[0015] At this time, a line laser scanner mounted on the end effector of a six-axis industrial robot is used to perform high-frequency dynamic scanning of the surface of the A-type car bumper according to a preset spatial grid trajectory. During the scanning process, the line laser sensor projects line laser structured light onto the surface of the bumper, and a high-resolution industrial camera captures the deformed light stripe after the laser is highly modulated on the surface of the object. The system calculates the three-dimensional spatial coordinates of the center pixel of the light stripe based on the principle of triangulation, thereby obtaining high-precision three-dimensional point cloud data containing density and depth information.

[0016] The acquired raw point cloud data is preprocessed, including statistical filtering to remove outliers and voxelization to reduce data redundancy. Point cloud registration is performed using a kd-tree-accelerated nearest neighbor search algorithm to unify the multi-view scan data into a global coordinate system. Based on this, a greedy projection triangulation algorithm is used to topologically connect the discrete point clouds according to the consistency of the point normal vectors, reconstructing a high-fidelity 3D mesh model of the area to be processed. This model not only contains the geometric shape information of the bumper surface, but also implicitly contains the topological connection relationship of the surface.

[0017] After obtaining the 3D mesh model, a local neighborhood geometric analysis window is constructed for each vertex Vi on the model. To accurately calculate the differential geometric properties, a quadratic surface fitting strategy is adopted: within the k-nearest neighbor range of vertex Vi, a locally parameterized parabolic surface S(u,v)=au²+bv²+cuv+du+ev+f is fitted using the least squares method. This parabolic surface can approximately represent the surface morphology of the local region. Based on the first and second fundamental forms of surface theory, the principal curvatures κ₁ and κ₂ of the fitted surface are calculated. 2. That is, the maximum and minimum values ​​of the normal curvature at the vertex, and the characteristic direction corresponding to the principal curvature is marked as the principal direction; on this basis, the Gaussian curvature K=κ1×κ2 of the vertex is calculated by multiplying the principal curvatures, and the average curvature H=(κ1+κ2) / 2H=(κ1+κ2) / 2 is calculated by using the average value of the principal curvatures; the system traverses all vertices of the 3D mesh model, and binds and stores the calculated κ1, κ2 and K values ​​as attribute data in the corresponding mesh vertices, thereby forming a continuously distributed curvature feature field on the model surface.

[0018] Based on the curvature attributes of the markers, the system performs continuous surface segmentation on the region to be processed. The system defines a curvature change rate evaluation index, specifically calculated by measuring the gradient magnitude of the principal curvature vectors between adjacent vertices. The algorithm is implemented using κ; it tracks the curvature change trend along the isoparametric lines of the mesh, and if the absolute value of the principal curvature difference between adjacent vertices in a certain region is |κ1(i+1)... κ1(i)∣ and∣κ2(i+1) If κ2(i)| is less than the preset first threshold δ1, and the sign of the Gaussian curvature K does not change from positive to negative (i.e., it does not cross the navel or flat point), then the region is determined to have curvature continuity. At the same time, the second derivative of the curvature along the tangent direction is calculated. If the curvature derivative is continuous, non-continuous regions where the curvature changes sharply are eliminated. Through the region growing algorithm, the mesh patches that satisfy the curvature change rate continuity constraint are aggregated to identify smooth, continuous surface patches without abrupt changes. Based on the spatial distribution and area size of each continuous surface patch, the system selects the processing areas that meet the requirements of multi-level grinding process and excludes irregular small feature areas.

[0019] For each identified continuous surface patch, its global geometric feature vector is extracted to describe its spatial orientation and shape. The feature vector construction process is as follows: the geometric center point Pcenter of the continuous surface patch is calculated as the spatial positioning reference; secondly, based on the normal vector Ni of all vertices in the surface patch, the average normal vector Navg of the surface patch is calculated by weighted averaging, which determines the initial approach direction of the grinding tool.

[0020] Principal Component Analysis (PCA) is used to process the 3D coordinate point set of the surface patch, and the eigenvector corresponding to the largest eigenvalue is extracted as the principal texture direction Dprincipal of the surface. This direction is used to plan the tool path direction for subsequent grinding. The maximum span distance Lspan, minimum radius of curvature Rmin, and average radius of curvature Ravg of the surface patch are used as geometric attribute parameters. The extracted geometric feature vector is expressed as F=[Pcenter,Navg,Dprincipal,Lspan,Rmin,Ravg]. This vector completely defines the geometric features of the continuous surface and provides accurate data input for the subsequent generation of grinding trajectories that highly match the geometric characteristics of the surface.

[0021] Specifically, the car bumper is the rear bumper of a certain sports sedan, and its surface contains complex feature shapes; the continuous curved surface specifically refers to the large arc transition surface that runs through both sides of the main body of the bumper. This area connects the side of the car body and the rear license plate area, and is a highly visible area of ​​the entire vehicle's appearance, so its surface quality requirements are extremely high.

[0022] A line laser scanner performs a full-field scan of the car bumper. Because the continuous curved surface is located in the center of the bumper, the point cloud data acquired in this area exhibits a smooth, dense distribution with no obvious data gaps. After reconstructing the 3D mesh model, the system calculates the curvature of the continuous curved surface region. The calculation results show that the principal curvatures κ1 and κ2 at each point in this region are positive, exhibiting elliptic point characteristics, and the Gaussian curvature K remains consistently positive with minimal fluctuations. More importantly, within the continuous curved surface region, the gradient magnitude of the rate of change of curvature tracked along the scanning direction is... κ remained at 0.05 mm. 2 The values ​​below are far below the preset threshold δ1, indicating that the surface of the region is extremely smooth and there are no discontinuous features such as edges or ribs.

[0023] Based on the continuity analysis of this rate of curvature change, the system successfully segmented and identified the continuous surface from discontinuous areas such as the radar hole area and taillight mounting slot on the bumper surface. In the feature extraction stage, the system calculated that the geometric center point of the continuous surface is located on the symmetry axis of the bumper, and its average normal vector Navg points roughly upward and backward, for example, at a 45-degree angle to the ground, with the main texture direction Dprincipal distributed along the transverse direction of the vehicle body. These extracted geometric feature vectors will be directly used to guide the subsequent step S112, ensuring that the generated grinding path can perfectly fit the large arc geometry of the continuous surface and provide accurate normal basis for subsequent rough grinding, semi-fine grinding and fine grinding.

[0024] Furthermore, based on the feature recognition of the continuous surface, the corresponding geometric feature vector is determined, and the corresponding initial tool path is generated by combining the adaptive path planning mechanism. The initial tool path uses B-spline curves to smooth the transition in the curvature change region of the continuous surface to avoid motion impact caused by path inflection points. This takes into account the overall consideration of the continuous surface and ensures the accuracy of the corresponding geometric feature vector.

[0025] At this point, the system reads the geometric feature vector F=[Pcenter,Navg,Dprincipal,Lspan,Rmin,Ravg] of the continuous surface extracted in step S111; based on the main texture direction Dprincipal, the overall tool path direction of the initial path is determined, usually along the curvature streamline direction or the material texture direction to maximize surface uniformity; at the same time, combined with the minimum curvature radius Rmin and the maximum span distance Lspan, the row spacing of the path is calculated. The row spacing setting must ensure that the residual height between adjacent tool trajectories meets the machining tolerance requirements; according to the surface boundary constraints, the system generates an initial path grid covering the entire area to be machined in the parameter domain of the continuous surface. This path grid consists of a series of parallel or isoparametrically distributed initial trajectory segments, each of which is mapped back to three-dimensional Cartesian space, forming a preliminary spatial path point set.

[0026] After generating the initial path point set, the system performs secondary detection of local geometric properties for each discrete tool position on the path. Specifically, it calculates the rate of change of the angle between adjacent tool positions and the deflection gradient of the normal vector. If the angle Δθ between adjacent tangent vectors on a certain path segment exceeds the preset smoothing threshold α, or the path curvature κpath of the path segment exceeds the curvature threshold κmax corresponding to the maximum allowable angular velocity constraint of the robot end effector, then the region is determined to be a curvature abrupt change region. This abrupt change usually originates from the transition zone between a continuous surface and other geometric features (such as chamfers and edges), or from the sharp bends of the surface itself. The system marks the starting point, ending point, and curvature extreme point of such curvature abrupt change regions as key control points, and assigns normal constraints and position coordinates to each key point in preparation for subsequent reconstruction processing.

[0027] For the identified curvature abrupt change regions, a non-uniform rational B-spline (NURBS) curve algorithm is used to locally reconstruct the path. The system uses marked key control points as shape points. While keeping the start and end positions of the path unchanged, a control vertex adjustment strategy is introduced. By calculating the basis functions of the B-spline, at least third-order continuous curve segments are generated to replace the broken-line inflection points in the original initial path. During this process, the algorithm must satisfy the C2 continuity condition, i.e., curvature continuity, to ensure that the generated curves do not have abrupt changes at the connection points. The core of the smoothing process is to straighten the acceleration components of the path. By calculating the arc length parameterization of the curve, it is ensured that the rate of change of the centripetal acceleration of the tool is within the linear response range of the robot's dynamic characteristics when passing through the abrupt change region. At the same time, for non-abrupt change regions, the original high-efficiency straight line or circular arc interpolation path is maintained, thereby maintaining overall machining efficiency while ensuring smoothness.

[0028] The system performs physical constraint verification on the complete initial toolpath after B-spline smoothing. The verification includes checking whether the vector change of the tool axis exceeds the universal joint angle limit, verifying whether the radius of curvature of the path is less than the minimum tool radius, and simulating whether the joint jerk of the robot moving along the path exceeds the limit. If the verification passes, the system generates the final initial toolpath data containing position coordinates, tool axis vector, feed rate and interpolation type. This path data is dynamically adaptive and can automatically adjust its smoothness according to the geometric changes of the surface, providing a reliable trajectory input for the precise motion control of the grinding robot in subsequent steps.

[0029] Specifically, the system uses the geometric feature vector F of the continuous surface to determine the main texture direction Dprincipal as the direction along the transverse side of the vehicle body, that is, from left to right through the rear of the vehicle, and uses this as the main tool path direction for rough grinding and fine grinding. Since the width of the continuous surface is relatively large, the system calculates an appropriate line spacing to cover the entire arc surface. However, at both ends of the continuous surface, that is, at the position connecting the side of the vehicle body, the geometry of the bumper shrinks sharply from the gentle large arc surface and transitions to the wheel arch feature area on the side.

[0030] When generating the initial toolpath, the system detects that there are obvious inflection points, i.e. curvature change areas, in these edge transition areas. If no treatment is done, the grinding robot will generate huge inertial forces when it moves at high speed through these inflection points, causing the tool to vibrate or even cut into the plastic substrate at key parts of the continuous curved surface, damaging the surface quality of the high-visibility area.

[0031] The system performs B-spline smoothing transition processing on the path inflection points at the two ends of the continuous surface. The system reconstructs the originally sharp polygonal path into a smooth and flowing third-order B-spline curve, which achieves continuous curvature connection with the straight path in the middle part. After smoothing, the end effector of the grinding robot shows a gradual change in speed and acceleration when entering and leaving the transition zone at the edge of the continuous surface, completely eliminating the mechanical impact caused by abrupt path changes. This processing ensures that the contact pressure of the grinding tool is constant in this highly visible area of ​​the continuous surface, thereby avoiding surface ripples and tool marks caused by pauses or impacts.

[0032] Therefore, based on the geometric parameters, material properties, and expected material removal rate of the grinding tool, the contact interference between the grinding tool envelope and the surface to be machined is determined, thereby determining the final grinding trajectory. This grinding trajectory includes the tool axis vector, step distance, and line distance, taking into account the overall considerations of the grinding tool's geometric parameters, material properties, and expectations, ensuring the accuracy of the final grinding trajectory.

[0033] At this point, the system constructs a contact interference calculation model based on the improved Preston equation. Inputs are the "geometric parameters of the grinding tool, such as the diameter of the abrasive belt wheel, the width of the contact wheel, and the thickness of the flap wheel," "material properties, such as the type of abrasive grain, the abrasive grain size, and the matrix hardness," and the expected material removal rate. Based on Hertzian contact theory, the system calculates the theoretical deformation of the grinding tool pressed into the surface of the car bumper under a preset normal force, i.e., the contact interference. This interference depends not only on the rigid geometric envelope of the tool but also on an elastic compensation coefficient determined by the tool's elastic modulus and the bumper material's springback characteristics. Through iterative solution of the equation, it determines that when the interference reaches a specific value, the material removal volume per unit stroke is exactly equal to the expected removal rate required by the process. This transforms the abstract process parameters into a specific geometric position offset, i.e., the distance the tool center needs to offset along the surface normal.

[0034] Based on the determined contact interference, the system performs attitude calculations for each tool position on the initial path. The tool axis vector is defined as the direction vector of the tool's central axis in space, and its determination must follow the "minimum tilt angle principle" and the "interference avoidance principle." At this time, in the flat area of ​​the continuous curved surface, the tool axis vector is set to be collinear and coincident with the surface normal vector at that point to ensure the orthogonal cutting effect of the abrasive grains. In the area with greater curvature, the system calculates the minimum tilt angle required to avoid collision with the non-machined surface based on the geometric envelope shape of the tool, and performs Euler angle transformation on the tool axis vector. In this way, a tool axis vector sequence that dynamically changes with the geometric features of the curved surface is generated to ensure that the tool envelope surface always fits the surface to be machined in the correct attitude and leaves a safety avoidance margin.

[0035] Based on the expected surface roughness requirements, the system performs final calculations on the path step (PathStep, the distance between adjacent interpolation points on the same path) and path interval (PathInterval, the distance between two adjacent parallel trajectories). For the path interval, the system combines tool geometry parameters (such as the effective radius of the grinding head) and contact interference, and uses the residual height model formula of ball end mills or flat end mills to inversely calculate the maximum path interval value that satisfies the maximum allowable residual height condition. For the step, the system adaptively adjusts it according to the local curvature changes of the surface: in areas with large curvature, the step is reduced to increase the density of discrete points and ensure fitting accuracy; in flat areas, the step is increased to improve computational efficiency. The final generated trajectory point set achieves an optimal balance of data volume while ensuring machining accuracy.

[0036] The system integrates the calculated contact interference, tool axis vector sequence, step distance, and row spacing information to generate the final grinding trajectory data structure. This structure includes not only spatial position information (x, y, z) and attitude information (i, j, k), but also the normal compensation value corresponding to each trajectory point. The system performs a comprehensive verification of the final trajectory, simulating the envelope of the tool moving along the trajectory to check for "overcutting: cutting too deep" or "undercutting: not contacting the surface" phenomena, and verifies whether there are unprocessed areas between the row spacings. If the verification passes, the system outputs the final grinding trajectory code that conforms to the robot controller communication protocol. This trajectory accurately guides the motion attitude and position of the grinding end effector in space.

[0037] Specifically, the system determines the trajectory for the rough grinding process of continuous curved surfaces; the selected grinding tool is a radial base flap wheel with a diameter of 60mm, the abrasive is alumina with a grit size of P80, and the expected material removal rate is set to remove the 0.15mm thickness required for the injection molding parting line; based on the flexibility of the flap wheel and the springback rate of the TPO bumper material, the system calculates that approximately 3.5mm of contact interference must be applied, meaning the center of the tool needs to be 3.5mm lower than the theoretical surface to overcome material springback and generate sufficient cutting force to achieve the expected removal rate.

[0038] When determining the cutter axis vector, considering the abrupt change in curvature at the connection points of the continuous curved surface and the side of the vehicle body, the system automatically deflects the cutter axis vector at these two locations outward by 15 degrees relative to the surface normal to avoid rigid interference between the edge of the metal center disc of the flap wheel and the side of the bumper, thus forming a "dragging" posture, while maintaining the normal posture in the flat area in the middle.

[0039] Since continuous curved surfaces are highly visible areas, the uniformity of surface texture is extremely important. Based on the interference amount of 3.5mm and the radius of the flap wheel, the system calculates and strictly sets the row spacing of two adjacent grinding paths to 8mm, which is about 13% of the tool diameter. This ensures that the residual height at the intersection of the two paths is less than 5 micrometers, completely eliminating the missed grinding area. The final output grinding path accurately guides the flap wheel to conform to the complex curvature of the continuous curved surface. By controlling the interference amount and posture, the efficient removal of the parting line is achieved, while ensuring a perfect transition of surface curvature.

[0040] refer to Figure 3 In step S12, the specific steps are as follows: S121: Based on the gradient change requirements of surface roughness and material removal depth, the grinding trajectory is divided into a rough grinding section, a semi-fine grinding section, and a fine grinding section in the time series. The rough grinding section uses a first-level grinding tool to quickly remove the machining allowance. The semi-fine grinding section uses a second-level grinding tool to eliminate the macroscopic ripples left by the rough grinding. The fine grinding section uses a third-level grinding tool to achieve a mirror or matte finish. S122: Obtain the preset process rule library, assign matching feed speed, spindle speed and contact pressure parameters to each grinding stage, and set the inter-stage tolerance coupling strategy, that is, the process tolerance range of the next grinding stage strictly covers and is less than the arithmetic mean deviation of the surface texture depth and contour left by the previous grinding stage, ensuring that the processing defects of the previous stage are completely removed in the next grinding stage without introducing new secondary surface damage.

[0041] In the embodiments of this application, the grinding trajectory is divided into a rough grinding section, a semi-fine grinding section and a fine grinding section in time sequence according to the gradient change requirements of surface roughness and material removal depth. The rough grinding section uses a first-level grinding tool to quickly remove the processing allowance, the semi-fine grinding section uses a second-level grinding tool to eliminate the macroscopic ripples left by the rough grinding, and the fine grinding section uses a third-level grinding tool to achieve a mirror or matte effect.

[0042] At this point, the system establishes a gradient quantification model of machining allowance and surface roughness based on the initial state and final target state of the surface to be processed. The system detects the initial defect depth of the area to be processed, such as injection weld lines and parting line height, to determine the total machining allowance. Simultaneously, based on the requirements of the final product for gloss or texture, a target roughness value is set. Then, a decreasing gradient curve is constructed between the total allowance and the target value: the rough grinding stage undertakes the largest proportion of material removal, with the goal of reducing surface defects to the level of macroscopic ripples; the semi-fine grinding stage undertakes a medium proportion of removal, focusing on eliminating cutting textures left by the previous process; the fine grinding stage undertakes minor removal or only surface finishing, focusing on improving the microscopic smoothness of the surface. Based on this model, the system calculates the tolerance zone range and feed time window to be allocated to each stage, providing a quantitative basis for the subsequent stage division.

[0043] At the beginning of the time series, the system defines a coarse grinding stage. The core objective of this stage is to efficiently remove material, so a longer processing time or a faster feed rate is allocated. The system is equipped with a first-level grinding tool, which uses a low-mesh, high-hardness abrasive matrix, such as diamond grinding wheels with a grit size of P60 to P80 or coarse-grained abrasive belts, which have extremely strong cutting ability and a large chip space. In terms of process parameters, the first-level tool is set to high depth of cut, high feed rate, and high contact pressure. The system sets the trajectory control points of this stage to a "large step size, low density" mode to maximize the material removal rate and quickly flatten the surface to be processed until the macroscopic surface undulations are reduced to the preset coarse grinding termination threshold.

[0044] Following the rough grinding stage, the system defines a semi-finish grinding stage. This stage, located between the end of rough grinding and the beginning of finish grinding, plays a crucial role in bridging the two stages. The system is equipped with a second-level grinding tool, which uses a medium-grit abrasive with a certain degree of flexibility, such as a flap wheel or abrasive belt with a grit size of P180 to P240. The cutting ability of the second-level tool is moderate, mainly used to trim the deep and wide tool marks left by the rough grinding stage. In terms of process parameters, the contact pressure is reduced compared to the rough grinding stage, the feed rate is moderately slowed down, and the row spacing is reduced to ensure that the peaks left by rough grinding are covered. The process logic of this stage is "mark removal and leveling", that is, to completely eliminate the macroscopic texture left by rough grinding, so that the surface roughness decreases by an order of magnitude, providing a uniform base surface for the final finish grinding.

[0045] At the end of the time series, the system defines a fine grinding stage, which occupies the final time window and is a crucial step in determining the final appearance quality. The system is equipped with a third-level grinding tool, which uses high-grit, high-elasticity abrasives or polishing media, such as fine sanding discs with a grit size of P400 to P600 or higher, wool polishing discs, or even ultra-fine grinding pads used for varnish treatment. The third-level tool has an extremely low cutting removal rate, but its micro-cutting edges are dense, and its main function is to perform micro-slipping and polishing on the surface. In terms of process parameters, the system sets extremely low contact pressure, extremely slow feed speed, and extremely small pass spacing, and even uses multi-pass reciprocating overlapping grinding. The goal of this stage is to eliminate the micro-scratches left by semi-fine grinding and achieve a high-gloss mirror effect or a uniform and delicate matte effect according to process requirements, ensuring that the surface roughness meets the final quality indicators.

[0046] Specifically, the system divides the multi-level grinding of the continuous surface into time series; the system detects that the continuous surface has an injection parting line with a height of about 0.5mm and a high initial surface roughness; in order to achieve high-quality matte coating pretreatment, the system constructs a gradient model from Ra12.5μm (initial) to Ra0.8μm (target).

[0047] In the early part of the time series, the system is set to rough grinding stage; the first-level P60 grit large-pore abrasive belt is selected, the spindle speed is set to high speed, and the feed rate is 300mm / s. The main task of this stage is heavy cutting, which quickly removes 0.45mm of machining allowance in about 15 seconds and grinds the protruding parting line completely flat. At this time, the continuous curved surface only has uniform and deep P60 abrasive texture.

[0048] The time series enters the semi-finishing stage; the system automatically switches to the second-level P240 grit conical flap wheel, reducing the feed speed to 200mm / s and decreasing the contact pressure; this stage mainly takes about 20 seconds and is specifically used to eliminate the deep groove texture left by rough grinding; the flexible characteristics of the flap wheel allow it to conform well to the curvature of the continuous curved surface, increasing the surface roughness to about Ra3.2μm, and the surface presents a smooth frosted state.

[0049] The time series enters the final fine grinding stage; the system switches to the third-level P600 grit flexible grinding disc, the feed speed is further reduced to 100mm / s, and the row overlap rate is increased to 60%; this stage takes about 25 seconds and mainly performs micro-cutting and surface finishing; after this stage, the micro-scratches on the continuous curved surface are completely eliminated, and the surface exhibits uniform, delicate and non-directional texture characteristics, which fully meets the stringent requirement of "surface quality before coating: Ra<0.8μm" for high visibility areas.

[0050] Furthermore, a preset process rule library is obtained, and matching feed speed, spindle speed and contact pressure parameters are assigned to each grinding stage. An inter-stage tolerance coupling strategy is set, that is, the process tolerance range of the next grinding stage strictly covers and is less than the arithmetic mean deviation of the surface texture depth and contour left by the previous grinding stage, ensuring that the processing defects of the previous stage are completely removed in the next grinding stage without introducing new secondary surface damage.

[0051] At this point, the system retrieves a preset process rule library from the storage medium. This database contains parameter combination matrices for "different material properties, such as TPO, ABS, PP+EPDM", "different tool types, such as abrasive belts, flap wheels, and wool wheels", and different surface morphologies. For the rough grinding section, semi-fine grinding section, and fine grinding section, the system retrieves the corresponding optimal process parameter triplet in the rule library based on the level label determined in step S121: feed rate Vf, spindle speed n, and contact pressure Fn.

[0052] At this point, for the rough grinding stage, the system maps and matches a combination of high feed rate, high spindle speed, and high contact pressure to maximize cutting efficiency; for the semi-finish grinding stage, it matches a medium feed rate, medium spindle speed, and medium-low contact pressure to balance cutting heat and surface smoothness; for the finish grinding stage, it matches a low feed rate, moderate or specific spindle speed, and low contact pressure to reduce surface micro-damage; the system binds these parameters to the corresponding trajectory segments on the time series to generate a process parameter package containing speed commands and force control commands.

[0053] After parameter allocation, the system constructs a mathematical model for the inter-level tolerance coupling strategy. The core of this strategy is to establish that the "lower limit of effective cutting depth" of the subsequent grinding stage must be greater than the "upper limit of surface micro-geometric features" left by the previous grinding stage. In specific implementation, the system calculates the maximum residual texture depth Rmax(prev) and the profile arithmetic mean deviation Ra(prev) generated by the previous grinding tool under specific process parameters. The system sets the process tolerance range for the subsequent grinding stage, requiring the minimum material removal layer thickness hmin(next) of the subsequent stage to satisfy the inequality: hmin(next) > k. Rmax(prev), where k is a safety factor, usually taken as 1.2 to 1.5, to ensure that even if there are processing errors, the peaks of the previous stage can be completely removed; at the same time, the surface roughness target value Ra(target) after the next stage of processing must be strictly smaller than the texture parameter of the previous stage, forming a unidirectional decreasing tolerance band.

[0054] To ensure that new secondary surface damage is not introduced while eliminating defects in the preceding stage, such as material melting due to overheating, plastic deformation due to indentation, or vibration marks caused by cutting vibration, the system performs preventative verification of the coupling strategy. Based on the thermophysical properties of the material, the system calculates the maximum temperature Tmax of the grinding zone and its corresponding heat-affected layer depth under the current combination of contact pressure and spindle speed. If the calculated heat-affected layer depth exceeds the tolerance coverage range set for the next stage, meaning that the thermally damaged layer cannot be completely removed by subsequent processes, the system triggers a correction mechanism to automatically reduce the contact pressure or spindle speed of the current stage until the heat-affected layer depth is within the tolerance coverage range. In addition, the system verifies whether the cutting force causes the workpiece to undergo elastic deformation beyond the allowable range, thereby avoiding contour errors caused by the "tool deflection" phenomenon and ensuring the consistency of geometric accuracy transferred between stages.

[0055] Specifically, the system retrieves parameters from the process rule library for TPO materials; for the rough grinding section of continuous curved surfaces, P60 abrasive belts are used, and the system automatically matches the feed speed to 300 mm / s, the spindle speed to 3000 rpm, and the contact pressure to 45 N; according to calculations, under this parameter combination, the P60 abrasive belt will leave a uniform grinding texture with a depth of about 50 micrometers on the continuous curved surface.

[0056] To ensure that the semi-finish grinding stage can completely eliminate these textures, the system activates an inter-stage tolerance coupling strategy. The setting of "semi-finish grinding stage: using P240 flap wheel" must meet the requirement that its minimum cutting depth is greater than 50 micrometers. Therefore, the system allocates a feed rate of 150 mm / s, a spindle speed of 4500 rpm, and a contact pressure of 25 N to the semi-finish grinding stage, thereby ensuring that a removal layer thickness of at least 60 micrometers is generated, and setting the target value of surface roughness Ra after semi-finish grinding to be controlled below 3.2 micrometers, which is strictly less than the Ra value left by rough grinding, approximately 12.5 micrometers.

[0057] Furthermore, for the "fine grinding stage: using a P600 polishing wheel," to prevent secondary damage such as "surface charring" caused by frictional heat in the TPO material due to excessive pressure, the system underwent preventative verification. The system calculations revealed that if the pressure of the semi-fine grinding was used, the thinner areas at the edges of the continuous curved surface might accumulate heat exceeding the material's softening point. Therefore, the system automatically triggered a correction mechanism, reducing the contact pressure of the fine grinding stage to 8N, while simultaneously reducing the feed rate to 80mm / s and increasing the spindle speed to 6000rpm. This was done to remove the micro-textures left by the semi-fine grinding using gentle, high-speed polishing. This strategy ensured that the highly visible areas of the continuous curved surface not only eliminated all previous textures but also did not produce any thermal damage or indentations, achieving a perfect mirror-like matte finish.

[0058] refer to Figure 4 In step S13, the specific steps are as follows: S131: The end effector of the grinding robot integrates a torque sensor. The end effector of the grinding robot performs grinding operations according to the combination of process parameters and monitors the actual contact force between the grinding tool and the surface of the bumper in real time during the movement along the grinding trajectory. The actual contact force is compared with the preset target contact force in real time, and the corresponding deviation is determined in combination with the force-position hybrid controller. The deviation includes the contact force deviation and its rate of change. S132: Based on the identification of deviation content, multiple sub-deviation items are determined, and the corresponding dynamic compensation command is determined in combination with the dynamic compensation mechanism of the grinding robot. Based on the response of the dynamic compensation command, the feed depth of the grinding tool in the normal direction is adjusted in real time, and the attitude angle of the grinding tool is adjusted at the same time. The dynamic fine adjustment of the attitude angle ensures that the contact line speed between the cutting edge and the surface of the continuous curved surface is within the preset speed range.

[0059] In the embodiments of this application, the end effector of the grinding robot integrates a torque sensor. The end effector of the grinding robot performs grinding operations according to the combination of process parameters and monitors the actual contact force between the grinding tool and the surface of the bumper in real time during the movement along the grinding trajectory. The actual contact force is compared with the preset target contact force in real time, and the corresponding deviation is determined in combination with the force-position hybrid controller. The deviation includes the contact force deviation and its rate of change, and the contact force deviation and its rate of change are introduced.

[0060] At this point, the six-dimensional torque sensor integrated between the end effector and the grinding tool of the grinding robot begins to work, acquiring the three-dimensional force components (Fx, Fy, Fz) and three-dimensional torque components (Mx, My, Mz) in the Cartesian coordinate system in real time at a sampling frequency higher than 1kHz. Since the grinding tool itself has mass, and the robot is affected by the gravity component in different postures, the system needs to have a built-in gravity compensation algorithm and inertial force compensation algorithm. Based on the real-time feedback of the robot's current joint angle, combined with the link dynamic parameters of the end effector, such as mass, center of mass position, and moment of inertia, the algorithm calculates the virtual force components generated by the tool's own weight and motion, and removes them from the raw data collected by the sensor. After decoupling and filtering, the system extracts the normal force data perpendicular to the tangent plane of the continuous curved surface, which is regarded as the actual contact force between the grinding tool and the bumper surface.

[0061] The system uses the target contact force Ftarget as the reference input signal and performs a cycle-by-cycle dynamic comparison with the actual contact force Factual. Considering that the grinding process is dynamic and continuous, the system not only compares the absolute value of the instantaneous force, but also introduces a time window moving average mechanism to eliminate high-frequency measurement noise. The comparison process calculates the instantaneous force error e(t) = Factual(t). Ftarget(t); Meanwhile, in order to capture the fluctuation trend during the grinding process, the system also monitors the integral term of the force error to assess whether there is a continuous overcut or undercut phenomenon; In different curvature regions of the continuous surface, the target contact force may be adaptively adjusted according to the trajectory curvature, so the comparison process needs to call the trajectory index synchronously to ensure that the actual force is accurately matched with the target setting value at the corresponding position.

[0062] To achieve precise control over grinding quality, the system employs a force-position hybrid control architecture to comprehensively calculate deviations. This controller decomposes the degrees of freedom in the operating space into position control direction and force control direction. In the force control direction (typically the surface normal), the system determines detailed deviations, including not only the contact force deviation e(t) mentioned above, but also the rate of change of the contact force de(t) / dt, i.e., the dynamic error rate of the force, calculated through differential operations. Combined with the force-position hybrid control strategy, the system maps this multidimensional deviation into position correction and impedance parameter adjustment. At this point, the contact force deviation determines the direction and magnitude of compensation for the normal feed depth, while the rate of change of the force reflects the matching degree between the system's rigidity and dynamic response, used to adjust the controller's damping ratio and gain. The determined deviations ultimately form a set of deviation vectors containing a force error scalar, an error rate vector, and corresponding correction coefficients, serving as the input for the next stage of the dynamic adjustment algorithm.

[0063] Specifically, the grinding robot carries a P240-grit flap wheel to perform grinding operations along a continuous curved surface. Since the continuous curved surface is not absolutely rigid and has complex hyperbolic changes, the torque sensor detects the interaction between the tool and the TPO plastic surface in real time during the robot's movement.

[0064] For example, when the robot end effector moves to the flat central region of a continuous curved surface, the sensor collects an initial normal force reading of 45N. After the system calculates using a gravity compensation algorithm and deducts the 5N gravity component of the end effector in that posture, the actual contact force Factual is 40N. At this point, the system compares this with the preset "target contact force Ftarget: set to 25N" for that region and calculates a contact force deviation e(t) of +15N, indicating a serious risk of overcutting.

[0065] Meanwhile, the force-position hybrid controller detects an extremely high rate of change of deviation de(t) / dt, indicating a sudden and sharp increase in contact force. Based on this, the system determines the deviation to be "positive out-of-tolerance with high dynamic impact," which is usually because the normal vector of the continuous surface changes faster than the robot's attitude response, causing the tool to "collide" with the surface. Based on this determined deviation, the system determines that a tool retraction action must be executed immediately and the normal stiffness reduced, thus providing precise control instructions for the dynamic adjustment in step S132. Conversely, when the robot moves to a thinner area at the edge of the continuous surface, if the actual contact force is detected to be lower than the target value and the rate of change is negative, the system determines it to be "undercut and false contact," and similarly generates a corresponding deviation vector for correction.

[0066] Furthermore, based on the identification of deviation content, multiple sub-deviation items are determined, and the corresponding dynamic compensation command is determined in combination with the dynamic compensation mechanism of the grinding robot. Based on the response of the dynamic compensation command, the feed depth of the grinding tool in the normal direction is adjusted in real time, and the attitude angle of the grinding tool is adjusted at the same time. The dynamic fine adjustment of the attitude angle ensures that the contact line velocity between the cutting edge and the surface of the continuous curved surface is within the preset speed range, which takes into account the overall deviation content and ensures the accuracy of multiple sub-deviation items.

[0067] At this point, the system performs decoupling analysis on the deviation content determined in step S131, identifying the specific sub-deviation items that constitute the total deviation. These sub-deviation items include: normal force steady-state error (Estatic), reflecting the static deviation between the current feed depth and the ideal value; normal force dynamic rate of change error (Edynamic), reflecting the inertial impact or following lag caused by sudden changes in surface curvature; and contact point tangential velocity deviation (Evelocity), reflecting the deviation of the actual cutting line velocity from the preset value due to changes in tool posture. Based on a preset weight matrix, the system performs weighted quantization calculations on the above sub-deviation items, converting them into correction quantities with physical meaning. For example, the normal force steady-state error is mapped to the displacement amount (ΔZ) that needs to be compensated, and the dynamic rate of change error is mapped to the acceleration correction value that needs to be adjusted, thereby providing accurate quantitative input for subsequent control strategies.

[0068] Based on the identified sub-deviation items, the system invokes the dynamic compensation mechanism of the grinding robot. Combining admittance control or impedance control models, it generates corresponding dynamic compensation commands. The core of these compensation commands is to correct the positional offset of the robot's end effector in the normal direction of the curved surface. At this time, when the "normal force steady-state error is positive, i.e., the actual contact force is too large," the system generates a negative compensation command, controlling the robot's end effector to retract outward along the normal direction of the curved surface, reducing the feed depth (Δd). Conversely, when the "error is negative, i.e., the contact force is insufficient," a positive compensation command is generated, increasing the feed depth. This adjustment process is high-frequency and real-time, and the compensation amplitude is proportional to the magnitude of the deviation. Simultaneously, it is combined with an integral control loop to eliminate steady-state errors. Through this real-time closed-loop adjustment of the normal feed depth, the system can overcome the effects of uneven bumper material and clamping deformation, ensuring that the actual cutting depth is always maintained within the process tolerance zone.

[0069] While adjusting the feed depth, the system simultaneously executes a dynamic fine-tuning strategy for the attitude angle. The system calculates the position of the current cutting point on the tool envelope in real time and monitors the angle between the tool axis and the surface cutting plane, i.e., the pitch angle or tilt angle. To compensate for changes in the cutting radius caused by changes in feed depth or surface curvature, the system performs reverse correction of the attitude angle based on the linear velocity formula "V=ω×R, where ω is the main spindle angular velocity and R is the actual cutting radius". When an increase in feed depth leads to an increase in the actual cutting radius, which may cause the linear velocity to exceed the preset range, the system fine-tunes the tool attitude angle, changes the contact area between the tool and the surface, and uses the speed difference of non-circular cutting points or changes the effective cutting radius to stabilize the contact linear velocity within the preset range. In addition, the fine-tuning of the attitude angle can also optimize the tool entry angle, avoid tool edge chipping or interference at abrupt changes in curvature of continuous surfaces, and ensure the smoothness of the machining process.

[0070] Specifically, the robot is performing semi-finish grinding on a continuous curved surface. When the robot reaches the large-curvature concave area where the continuous curved surface connects to the side panel of the vehicle body, the torque sensor detects that the actual contact force instantly spikes to 15N above the target value, thus identifying a deviation in the positive normal force. Based on this, the system immediately triggers a dynamic compensation mechanism; it calculates the required retraction of 0.8mm based on the magnitude of the deviation, generates a normal compensation command, and controls the robot's end effector to rapidly retreat along the local normal direction of that point on the continuous curved surface, reducing the feed depth and causing the contact force to quickly drop back to a stable range of 25N.

[0071] Meanwhile, due to the large curvature of this region, in order to maintain a constant cutting speed and avoid the phenomenon of "over-grinding and whitening" on the surface, the system dynamically fine-tunes the attitude angle. As the feed depth decreases, the actual cutting point moves towards the edge of the flap wheel, resulting in a slight increase in the cutting radius and an upward trend in the linear speed. Therefore, the system generates an attitude angle correction command, controlling the end effector to deflect inward by about 2 degrees relative to the surface normal, so that the contact point moves back towards the center of the flap wheel. This utilizes the lower linear speed at the center point to offset the spindle fluctuations, ensuring that the contact linear speed between the cutting edge and the continuous curved surface is strictly locked within the preset range of 22m / s to 25m / s. Through this series of coordinated adjustments of feed depth and attitude angle, the robot successfully passes through the high curvature region, and the surface quality of the continuous curved surface is uniform and consistent, without any over-cutting or burning marks.

[0072] refer to Figure 5 In step S14, the specific steps are as follows: S141: During the grinding process of the grinding tool on the continuous curved surface, the multi-physics field signal of the continuous curved surface is collected simultaneously, and the image data of the processed area captured by the machine vision system is also collected. At this time, the vibration signal and temperature signal are determined based on the recognition of the multi-physics field signal, and the corresponding grinding quality evaluation matrix is ​​constructed by combining the texture highlight features and shadow features in the image data. S142: Based on the identification of the grinding quality evaluation matrix, the corresponding waviness index is determined. The waviness index is compared with the preset index threshold. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, the online compensation mechanism is triggered. During the compensation process, the corresponding correction content combination is determined, thereby automatically correcting the trajectory offset of the subsequent unprocessed area, dynamically adjusting the contact pressure parameters, synchronously marking the grinding process of the continuous surface, and gradually completing the multi-level grinding of the entire continuous surface.

[0073] In the embodiments of this application, during the grinding process of the grinding tool on the continuous curved surface, multi-physics field signals of the continuous curved surface are collected simultaneously, and image data of the processed area captured by the machine vision system are also collected. At this time, vibration signals and temperature signals are determined based on the recognition of the multi-physics field signals, and a corresponding grinding quality evaluation matrix is ​​constructed by combining the texture highlight features and shadow features in the image data. This approach takes into account the overall consideration of multi-physics field signals and ensures the accuracy of vibration signals and temperature signals.

[0074] During the movement of the grinding tool along the continuous curved surface, the system synchronously acquires multi-physics field signals through an integrated high-sensitivity sensor network. Specifically, this includes: acquiring triaxial vibration signals using piezoelectric accelerometers installed near the end effector or spindle, which reflects the dynamic excitation, tool imbalance, and modal vibration of the workpiece-tool system during abrasive cutting; acquiring temperature signals of the grinding zone using infrared thermal imagers or embedded miniature thermocouples, which reflects the severity of frictional heat generation and the risk of material thermal damage; the acquisition process must strictly follow the time synchronization protocol to accurately align the timestamps of the physical signals with the robot's spatial position information; the system preprocesses the raw signals, applying a bandpass filter to remove high-frequency electromagnetic noise and low-frequency gravity components from the vibration signals, and performing drift compensation and smoothing on the temperature signals, thereby extracting a feature signal sequence that truly reflects the machining state.

[0075] Meanwhile, a machine vision system mounted on the robot's end effector or external fixed bracket is used to image the freshly polished area in real time. To capture the subtle textures on the bumper surface, the system employs a specific structured light illumination scheme, such as multi-angle LED dome light or strip light, to induce specific highlight and shadow effects on the smooth surface. The camera acquires two-dimensional image data of the processed area and transmits it to the image processing unit. The system extracts the region of interest (ROI) from the image and uses image enhancement algorithms to improve contrast. It extracts texture highlight and shadow features from the image: highlight features reflect the microscopic smoothness and gloss consistency of the surface; areas with concentrated highlights are usually relatively smooth, while diffuse highlights indicate the presence of ripples. Shadow features reflect the surface roughness and texture depth; deep and sharp shadows correspond to deeper cutting marks or unpolished pits.

[0076] The system fuses the physical signal features in the time domain with the image features in the spatial domain to construct a grinding quality evaluation matrix. Then, the system extracts "time-domain indicators, such as root mean square (RMS) and kurtosis" and "frequency-domain indicators, such as dominant frequency amplitude and harmonic energy ratio," from the vibration signal as dimensions for evaluating "process stability." It also extracts the maximum temperature rise and temperature gradient from the temperature signal as dimensions for evaluating "thermal damage risk." Furthermore, it extracts the uniformity of highlight distribution, shadow density, and texture direction consistency from the image data as dimensions for evaluating "surface morphology quality." The system maps these heterogeneous data onto a unified quantization scale to construct an M×N quality evaluation matrix, where M represents the sampling time or spatial location, and N represents the feature dimension. This matrix not only contains the numerical values ​​of individual indicators but also the correlation weights between indicators, thus forming a comprehensive digital description of the current grinding status.

[0077] Specifically, the robot is performing a fine grinding process on a continuous curved surface. At this time, the system initiates a multimodal perception process. The accelerometer integrated on the end effector detects that in a certain arc segment of the curved surface, the RMS value of the vibration signal suddenly increases to 1.5 times the preset threshold, and the main frequency shifts. This indicates that the grinding head may be experiencing chatter or contact instability. At the same time, the infrared thermal imager monitors that the temperature of the grinding zone in this area has risen by about 15°C, approaching the softening point risk threshold of the TPO material.

[0078] The linear scan camera located at the rear of the robot captured a high-resolution image of the area that had just been polished. Due to the requirement of a high-gloss matte finish for the continuous curved surface, the system activated the "low-angle light" illumination mode. The image processing unit analysis revealed that there were long, discontinuous highlight bands in the image of this area, accompanied by irregular dark shadow stripes between the highlight bands. Based on this, the system determined that the discontinuity of the highlight bands indicated the presence of tiny wavy lines on the surface, while the dark shadows indicated that the texture left by the previous stage (semi-finish polishing) was not completely covered.

[0079] Based on the above analysis, the system constructs a polishing quality evaluation matrix at this moment. In the matrix, the two dimensions of "vibration energy" and "texture waviness" are marked as red warning states, and "temperature gradient" is marked as yellow warning states. This matrix intuitively reflects that the current polishing process has caused the surface waviness to exceed the standard due to flutter, and is accompanied by a slight risk of heat accumulation, thus providing solid data support for triggering the online compensation mechanism in step S142.

[0080] Furthermore, based on the identification of the grinding quality evaluation matrix, the corresponding waviness index is determined. The waviness index is compared with a preset index threshold. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered. During the compensation process, the corresponding correction content combination is determined, thereby automatically correcting the trajectory offset of the subsequent unprocessed area and dynamically adjusting the contact pressure parameters. The grinding process of the continuous surface is marked synchronously, and the multi-level grinding of the entire continuous surface is gradually completed. The deviation content is further controlled. The grinding quality evaluation matrix and the online compensation mechanism are fully considered, realizing the multi-level grinding of the entire continuous surface and improving the accuracy of multi-level grinding of the continuous surface.

[0081] At this point, the system performs in-depth analysis on the grinding quality evaluation matrix constructed in step S141, focusing on extracting the waviness index that characterizes the micro-geometric undulations of the surface. The system uses a weighted fusion algorithm to map the high-frequency vibration components in the evaluation matrix into indirect indicators of surface micro-roughness, and directly converts the uniformity of highlight distribution and the density of shadow features extracted from the machine vision image into contour waviness (Winess, Wa) values. At this point, by calculating the second derivative of the image grayscale gradient or extracting texture energy within a specific spatial frequency range using frequency domain filtering, a real-time waviness value Wreal is constructed. This value comprehensively reflects the degree of periodic surface undulations caused by tool chatter, uneven feed, or uneven material hardness, and is a key quantitative basis for determining whether the current grinding quality meets the requirements of Grade A surface.

[0082] The system rigorously compares the real-time calculated waviness index Wreal with the preset threshold Wthreshold. This threshold is a critical value set according to the specific appearance requirements of the continuous surface, such as mirror or matte finish, and usually corresponds to the boundary of visually perceptible waviness or orange peel defects. If Wreal ≤ Wthreshold, the current processing state is deemed qualified, and the system maintains the existing process parameters and continues to operate. If Wreal > Wthreshold, the surface quality is deemed unqualified, and an online compensation mechanism is immediately triggered. This triggering logic has a hysteresis locking characteristic, that is, once multiple consecutive sampling points exceed the standard, the system will lock the compensation state to prevent frequent oscillations caused by instantaneous interference and ensure the stability and necessity of the compensation action.

[0083] Once the online compensation mechanism is triggered, the system determines the corresponding combination of corrections based on the magnitude and nature of the excessive waviness. This combination includes two core adjustment strategies: First, automatic correction of the trajectory offset; based on the physical law that waviness is proportional to the fluctuation of cutting force, the system calculates the normal position correction ΔL required to eliminate waviness; if the waviness is convex, a negative offset command is generated to control the robot's end effector to retract along the normal of the curved surface to reduce the actual cutting depth; conversely, if it is concave, a small feed is performed; Second, dynamic adjustment of contact pressure parameters; the system synchronously adjusts the force control parameters. If the waviness is determined to be caused by chatter due to excessive stiffness, the target contact pressure is reduced and the damping ratio is increased; if the waviness is determined to be caused by tool slippage due to insufficient cutting force, the contact pressure is appropriately increased; the system sends the above combination of corrections to the robot's motion controller and force control unit in real time to achieve a smooth transition and adaptive adjustment of the process parameters for the subsequent unprocessed area.

[0084] While performing compensation adjustments, the system simultaneously marks the grinding progress of the continuous surface. The system maintains a dynamically updated processing status map, dividing the surface into four states: "to be processed," "in processing," "processed to standard," and "processed and compensated." As the robot moves along the trajectory, the system updates the coordinates of the current tool position on the surface parameter domain in real time and records the completion status of each level (rough grinding, semi-fine grinding, and fine grinding). If a certain area still does not meet the standard after compensation, the system will mark the area and plan a secondary tool path. This process continues to loop until the system detects that all mesh units of the continuous surface have completed the corresponding level of grinding and the waviness index meets the threshold requirements, thus finally completing the multi-level grinding task of the entire continuous surface and outputting the final quality report.

[0085] Specifically, the system is currently in the fine grinding stage of the continuous curved surface, using a P600 polishing wheel. When processing a section of the curved surface near the left side of the rear license plate, the real-time waviness index Wreal calculated by the system based on the grinding quality evaluation matrix rises to 15μm, while the preset high visibility area index threshold Wthreshold is 5μm. Since 15>5, the system determines that visible fine flutter marks have appeared on the surface and immediately triggers the online compensation mechanism.

[0086] The system analyzes and corrects the combination of content, determining that the ripples originate from the local softness of the material in that area, causing the polishing wheel to sink too deeply and triggering high-frequency vibrations. Therefore, the system automatically corrects the offset of the subsequent unprocessed trajectory, controls the robot end effector to adjust outward by 0.2mm along the surface normal to reduce the envelope of the polishing wheel; at the same time, the target contact pressure is reduced from 8N ​​to 6N, and the damping coefficient of the force control system is increased to absorb residual vibrations.

[0087] As the robot continued to move to the right, these parameter adjustments took effect immediately, the surface ripples disappeared quickly, and the texture returned to uniformity. The system marked the arc surface as "processed and compensated" in the status map and continued to monitor the remaining area. The system confirmed that the entire multi-stage polishing task of the car bumper was completed when the continuous curved surface had passed the fine grinding process from left to right and no more exceeding the standard.

[0088] Please see Figure 6 The multi-stage polishing system for continuous curved surfaces in a car bumper is applied to the aforementioned multi-stage polishing method for continuous curved surfaces in a car bumper; the multi-stage polishing system for continuous curved surfaces in a car bumper includes: The grinding trajectory module 21 is used to collect the area to be processed of the car bumper, determine the continuous surface based on the identification of the area to be processed, determine the corresponding initial tool path based on the identification of the continuous surface, and determine the corresponding grinding trajectory in combination with the parameter combination of the grinding tool. The process parameter module 22 is used to divide the grinding trajectory into a rough grinding section, a semi-fine grinding section and a fine grinding section, and to assign corresponding grinding tools and process parameter combinations to the rough grinding section, the semi-fine grinding section and the fine grinding section respectively, and the process tolerance of the later grinding section covers the surface texture depth left by the previous grinding section. The deviation content module 23 is used for the end effector of the grinding robot to move along the grinding trajectory according to the combination of process parameters, and to monitor the actual contact force between the grinding tool and the surface of the bumper in real time during the grinding process, mark the deviation content between the actual contact force and the target contact force, and dynamically adjust the normal feed depth and attitude angle of the grinding head along the deviation content. The multi-stage grinding module 24 is used to collect vibration and temperature signals in real time during the grinding process, and to construct a grinding quality evaluation matrix by combining the image data of the processed area. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-stage grinding of the entire continuous surface is completed.

[0089] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; in fact, according to the embodiments of this disclosure, the features and functions of two or more modules or described above can be embodied in one module; conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

[0090] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein; this application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein; the specification and embodiments are to be considered exemplary only.

[0091] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-stage grinding method for continuous curved surfaces in an automobile bumper, characterized in that, include: The machined area of ​​the car bumper is collected, and a continuous surface is determined based on the identification of the machined area; the corresponding initial tool path is determined according to the identification of the continuous surface, and the corresponding grinding trajectory is determined by combining the parameter combination of the grinding tool. The polishing path is divided into coarse polishing, semi-fine polishing and fine polishing, and corresponding polishing tools and process parameter combinations are assigned to the coarse polishing, semi-fine polishing and fine polishing sections respectively. The process tolerance of the later polishing section covers the surface texture depth left by the previous polishing section. The end effector of the grinding robot moves along the grinding trajectory according to the combination of process parameters, and monitors the actual contact force between the grinding tool and the bumper surface in real time during the grinding process. It marks the deviation between the actual contact force and the target contact force, and dynamically adjusts the normal feed depth and attitude angle of the grinding head along the deviation. Vibration and temperature signals are collected in real time during the polishing process, and a polishing quality evaluation matrix is ​​constructed by combining the image data of the processed area. If the waviness index in the polishing quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-level polishing of the entire continuous surface is completed.

2. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 1, characterized in that, The process involves acquiring the area to be processed from the car bumper, identifying a continuous surface based on the identification of this area, determining the corresponding initial toolpath based on the identification of the continuous surface, and combining this with the parameter combination of the grinding tool to determine the corresponding grinding trajectory, including: A line laser scanner is used to dynamically scan the surface of a car bumper to collect three-dimensional point cloud data. A three-dimensional mesh model of the area to be processed is reconstructed based on the three-dimensional point cloud data. The principal curvature and Gaussian curvature of each point on the three-dimensional mesh model are marked. The corresponding continuous surface is identified based on the continuity of the rate of change of curvature.

3. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 2, characterized in that, The process involves acquiring the processing area of ​​the car bumper, determining a continuous surface based on the identification of the processing area, determining the corresponding initial toolpath based on the identification of the continuous surface, and determining the corresponding grinding trajectory by combining the parameter combination of the grinding tool. The process also includes: Based on the feature recognition of the continuous surface, the corresponding geometric feature vector is determined, and the corresponding initial tool path is generated by combining the adaptive path planning mechanism. The initial tool path uses B-spline curves to smooth the transition in the curvature change region of the continuous surface to avoid motion impact caused by path inflection points. Based on the geometric parameters of the grinding tool, material properties, and expected material removal rate, the contact interference between the grinding tool envelope and the surface to be machined is determined, thereby determining the final grinding trajectory, which includes the tool axis vector, step distance, and row distance.

4. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 1, characterized in that, The process involves dividing the polishing trajectory into coarse polishing, semi-fine polishing, and fine polishing stages, and assigning corresponding polishing tools and process parameter combinations to each stage. The process tolerance of each subsequent polishing stage covers the surface texture depth left by the preceding stage. Based on the gradient change requirements of surface roughness and material removal depth, the grinding trajectory is divided into a rough grinding section, a semi-fine grinding section, and a fine grinding section in time sequence. The rough grinding section uses a first-level grinding tool to quickly remove the machining allowance, the semi-fine grinding section uses a second-level grinding tool to eliminate the macroscopic ripples left by the rough grinding, and the fine grinding section uses a third-level grinding tool to achieve a mirror or matte finish.

5. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 4, characterized in that, The process of dividing the polishing trajectory into coarse polishing, semi-fine polishing, and fine polishing stages, and assigning corresponding polishing tools and process parameter combinations to each stage, with the process tolerance of each subsequent polishing stage covering the surface texture depth left by the previous stage, also includes: Obtain the preset process rule library, assign matching feed speed, spindle speed and contact pressure parameters to each grinding stage, and set the inter-stage tolerance coupling strategy, that is, the process tolerance range of the next grinding stage strictly covers and is less than the arithmetic mean deviation of the surface texture depth and contour left by the previous grinding stage, ensuring that the processing defects of the previous stage are completely removed in the next grinding stage without introducing new secondary surface damage.

6. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 1, characterized in that, The end effector of the grinding robot moves along the grinding trajectory according to the combination of process parameters, and monitors the actual contact force between the grinding tool and the bumper surface in real time during the grinding process. It marks the deviation between the actual contact force and the target contact force, and dynamically adjusts the normal feed depth and attitude angle of the grinding head along the deviation, including: The end effector of the grinding robot integrates a torque sensor. The end effector performs grinding operations based on a combination of process parameters and monitors the actual contact force between the grinding tool and the bumper surface in real time during its movement along the grinding trajectory. The actual contact force is compared with the preset target contact force in real time, and the corresponding deviation is determined in conjunction with the force-position hybrid controller. This deviation includes the contact force deviation and its rate of change.

7. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 6, characterized in that, The end effector of the grinding robot moves along the grinding trajectory according to a combination of process parameters, and monitors the actual contact force between the grinding tool and the bumper surface in real time during the grinding process. It marks the deviation between the actual contact force and the target contact force, and dynamically adjusts the normal feed depth and attitude angle of the grinding head along this deviation. The system also includes: Multiple sub-deviation items are determined based on the identification of deviation content, and the corresponding dynamic compensation command is determined in combination with the dynamic compensation mechanism of the grinding robot. Based on the response of the dynamic compensation command, the feed depth of the grinding tool in the normal direction is adjusted in real time, and the attitude angle of the grinding tool is adjusted at the same time. The dynamic fine adjustment of the attitude angle ensures that the contact line speed between the cutting edge and the surface of the continuous curved surface is within the preset speed range.

8. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 1, characterized in that, The process involves real-time acquisition of vibration and temperature signals during the polishing process, combined with image data of the processed area to construct a polishing quality evaluation matrix. If the waviness index corresponding to the polishing quality evaluation matrix exceeds a preset threshold, an online compensation mechanism is triggered, automatically correcting the trajectory offset of the subsequent unprocessed area and adjusting the contact pressure parameters until the multi-level polishing of the entire continuous surface is completed, including: During the grinding process of the grinding tool on the continuous curved surface, the multi-physics field signal of the continuous curved surface is collected simultaneously, and the image data of the processed area is collected by the machine vision system. At this time, the vibration signal and temperature signal are determined based on the recognition of the multi-physics field signal, and the corresponding grinding quality evaluation matrix is ​​constructed by combining the texture highlight features and shadow features in the image data.

9. The multi-stage grinding method for continuous curved surfaces in an automobile bumper according to claim 8, characterized in that, The process of acquiring vibration and temperature signals in real time during grinding, and constructing a grinding quality evaluation matrix by combining it with image data of the processed area; if the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-level grinding of the entire continuous curved surface is completed, also includes: The corresponding waviness index is determined based on the identification of the grinding quality evaluation matrix. The waviness index is compared with the preset index threshold. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, the online compensation mechanism is triggered. During the compensation process, the corresponding correction content combination is determined, thereby automatically correcting the trajectory offset of the subsequent unprocessed area, dynamically adjusting the contact pressure parameters, synchronously marking the grinding process of the continuous surface, and gradually completing the multi-level grinding of the entire continuous surface.

10. A multi-stage polishing system for continuous curved surfaces in an automobile bumper, characterized in that, The multi-stage polishing system for continuous curved surfaces in a car bumper is applied to the multi-stage polishing method for continuous curved surfaces in a car bumper as described in any one of claims 1-9; the multi-stage polishing system for continuous curved surfaces in a car bumper includes: The grinding trajectory module is used to acquire the area to be processed of the car bumper, determine the continuous surface based on the identification of the area to be processed, determine the corresponding initial tool path based on the identification of the continuous surface, and determine the corresponding grinding trajectory by combining the parameter combination of the grinding tool. The process parameter module is used to divide the grinding trajectory into rough grinding, semi-fine grinding and fine grinding sections, and assign corresponding grinding tools and process parameter combinations to the rough grinding, semi-fine grinding and fine grinding sections respectively. The process tolerance of the later grinding section covers the surface texture depth left by the previous grinding section. The deviation content module is used for the end effector of the grinding robot to move along the grinding trajectory according to the combination of process parameters, and to monitor the actual contact force between the grinding tool and the surface of the bumper in real time during the grinding process, mark the deviation content between the actual contact force and the target contact force, and dynamically adjust the normal feed depth and attitude angle of the grinding head along the deviation content. The multi-stage grinding module is used to collect vibration and temperature signals in real time during the grinding process, and to construct a grinding quality evaluation matrix by combining the image data of the processed area. If the waviness index corresponding to the grinding quality evaluation matrix exceeds the preset index threshold, an online compensation mechanism is triggered, and the trajectory offset of the subsequent unprocessed area is automatically corrected and the contact pressure parameters are adjusted until the multi-stage grinding of the entire continuous surface is completed.

Images