Microstructure adaptive polishing method based on point cloud topology perception

Abstract

This invention relates to the field of optical processing energy allocation technology, and discloses an adaptive polishing method for microstructures based on point cloud topology perception. The method includes: constructing a geometric topological curvature tensor field of the edge region using in-situ white light interferometric point cloud data and extracting geometrically invariant feature vectors; matching energy allocation parameters containing the output current amplitude of the power supply regulation circuit in a retrieval matrix using a local sensitive hash algorithm; determining the power load pulse width required for the power supply regulation circuit to drive the polishing load system; and controlling the polishing load system to run along a Peano fractal path. This invention compensates for the nonlinear distribution of edge pressure through precise energy allocation via the power supply regulation circuit, eliminates processing errors caused by boundary condition truncation, suppresses excessive material removal and the coherent superposition of residual processing textures, and effectively improves the conformal accuracy and surface quality of the edge region of precision optical components.

Description

Technical Field

[0001] This invention relates to an adaptive polishing method for microstructures based on point cloud topology sensing, belonging to the field of optical processing energy allocation technology. Background Technology

[0002] In current computer-controlled optical surface forming processes, material removal typically follows the Preston equation, which states that the amount of material removed is proportional to the processing pressure, speed, and dwell time. In this technical field, stable dwell time calculation algorithms are usually based on the physical premise that the tool influence function remains constant across the entire processing surface. However, for precision microstructured parts in power systems, this constant premise fails in the edge regions of the structure. When the flexible polishing head moves to the step edge of the microstructure, the reduced contact area causes a sudden increase in local pressure. This nonlinear pressure distribution leads to excessive removal of edge material and the formation of edge collapse defects, damaging the geometric integrity and surface electrical stability of the power micro-components. At the same time, conventional convolution algorithms produce boundary condition truncation when processing edge regions, causing the dwell time calculation results to deviate from the actual physical requirements, further aggravating the contour error. Although physical sacrificial block compensation is common in the processing of large parts, it is difficult to implement physical compensation within the dense microstructures of power distribution equipment due to space constraints.

[0003] To address edge effects, existing technologies attempt to break through at the tool structure level, but remain limited by the precision bottleneck of control strategies. Chinese invention patent CN112605786B discloses an airbag polishing device and its polishing head mechanism, employing a concentric ring structure of a main pressure airbag and side auxiliary airbags. It adjusts the internal pressure of each airbag in segments to balance the contact stress at the protruding edge of the polishing head. This method essentially relies on an open-loop control strategy based on preset parameters. Pressure adjustment is based on the overall geometric position of the workpiece edge, lacking real-time sensing of the surface morphology of the processing area. When dealing with heterogeneous microstructure arrays with continuously abrupt curvature changes, relying solely on the flexible passive adaptation of the airbags makes it difficult to respond to nanoscale topological changes, hindering precise contact pressure distribution and edge collapse correction capabilities. The existing technology has several limitations, including: 1. Nonlinear pressure distribution in edge-sensitive areas causes geometric collapse in microstructures; 2. Data truncation at boundaries prevents precise energy allocation for edge contours; 3. Conventional grating scanning paths induce residual mid-frequency ripples in edge regions, reducing the consistency of physical properties of precision surfaces. While in-situ detection methods such as white light interferometry exist, the industry lacks a control model capable of converting unstructured 3D point cloud data into deterministic edge protection strategies in real time, resulting in the processing yield of precision power components being limited by the complex surface topology.

[0004] Therefore, the technical problem to be solved by this invention is how to establish a virtual compensation model that adapts to spatial location based on the local topological features of in-situ feedback, and to combine it with a fractal path scheduling strategy to eliminate the geometric collapse and periodic texture residue at the edges of fine structures. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: an adaptive polishing method for fine structures based on point cloud topology awareness, comprising the following steps:

[0006] Step S1: Obtain white light interference point cloud data of the edge region of the microstructure array on the surface of the workpiece to be processed, and construct a curvature tensor field characterizing the local geometric topological boundary of the microstructure based on the white light interference point cloud data;

[0007] Step S2: Perform dimensionality reduction mapping on the curvature tensor field to extract feature vectors that characterize the spatial morphological properties of the microstructure. The feature vectors have rotation and translation invariance.

[0008] Step S3: Input the feature vector into the preset microstructure edge model association database, and perform an approximate nearest neighbor search in the hash space using the locality sensitive hash algorithm to obtain the matching energy allocation parameters. The energy allocation parameters include the output current adjustment amplitude of the power supply regulation circuit and the removal function kernel parameters used to limit the material removal efficiency of the microstructure unit.

[0009] Step S4: Calculate the power load pulse width required for the power supply regulation circuit to drive the polishing load system based on the energy allocation parameters. The power load pulse width corresponds to the dwell time of the polishing tool in the edge region. Generate the Peano fractal path based on the isotropic constraint conditions of the edge region.

[0010] Step S5: Based on the residence time, the power supply regulation circuit outputs a drive signal with amplitude modulation after output current regulation, which drives the polishing load system to perform material removal along the Peano fractal path. The duration of the drive signal is controlled by the pulse width to achieve electrical energy compensation for nonlinear pressure jumps in the edge region of the microstructure.

[0011] Preferably, step S1 specifically includes: scanning the microstructure array of the workpiece to be processed using an in-situ integrated white light interferometer to obtain a set of three-dimensional coordinate points with a sampling accuracy of not less than 1 nm; calculating the normal vector and principal curvature of each sampling point in the three-dimensional coordinate point set; constructing a curvature tensor field using the principal curvature H to achieve geometric topological characterization of the heterogeneous surface features in the microstructure array; the principal curvature H is determined by the following formula: ,in, The maximum curvature of the sampling point in the principal direction. The minimum curvature of the sampling point in the principal direction.

[0012] Preferably, in step S3, the microstructure edge model association database includes: a key field for storing preset geometric feature vector clusters, which are mapped to multiple hash buckets using a locality-sensitive hashing algorithm; and a value field for storing standard process model parameters that correspond one-to-one with the geometric feature vector clusters. The approximate nearest neighbor retrieval process follows the following mapping rules: ,in, To retrieve the obtained energy allocation parameters, Let f be the feature vector, and let f represent the distance metric function in the hash space.

[0013] Preferably, step S4 specifically includes: performing edge extension for the microstructure topological features in the algorithm space to construct a continuous metasurface to eliminate the boundary condition truncation caused when the polishing load system moves to the edge region; calling the target removal function from the preset nonlinear removal function library according to the tool suspension ratio of the polishing tool; and using the target removal function to correct the dwell time to suppress excessive removal of material in the edge region.

[0014] Preferably, in step S4, the irregular scanning of the Peano fractal path converts the regular mid-frequency ripple error generated in the edge region into high-frequency randomly distributed background noise, thereby reducing the spectral concentration of the surface morphology of the workpiece to be processed and optimizing the modulation transfer function.

[0015] Preferably, the method further includes: step S6, after the material removal action is completed, using white light interference point cloud data to detect processing errors; when the processing error exceeds a preset accuracy threshold, triggering a residual-driven adaptive correction mechanism to generate optimized parameter pairs.

[0016] Preferably, the adaptive correction mechanism performs the following actions through a background asynchronous computing thread: calculates the mapping residual between the processing error and the expected removal amount, and performs incremental iteration on the energy allocation parameters based on the mapping residual; updates the value range in the edge model association database to realize the evolution of process model parameters.

[0017] Preferably, the polishing load system includes an airbag polishing head. In step S5, the pressure distribution of the material removal action follows the Preston equation modified by the energy distribution parameter, and the pressure fluctuation range of the polishing load system is controlled between 0.01 MPa and 0.05 MPa.

[0018] Preferably, in step S2, a topological feature vector extraction algorithm is used to mask the coordinate differences of the microstructure array in absolute space in order to extract feature vectors.

[0019] Preferably, the method is used to process optical components containing heterogeneous microstructure arrays, and to increase the control parameter response frequency of the power supply regulation circuit to above 100Hz through near nearest neighbor search.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. In adaptive edge conformal polishing, a local topological coding mechanism is introduced to transform the acquired unstructured point cloud into a feature vector with geometric invariance. The approximate nearest neighbor search algorithm based on hash index is used to match the preset process model. This transforms the process of constructing control parameters for complex edges from traditional online high-order numerical iterative calculation to constant-level vector retrieval, eliminating the computational resource blockage caused when processing massive heterogeneous microstructure arrays and ensuring the real-time response performance of the control system under high-speed axis motion.

[0022] 2. A mathematically continuous metasurface is constructed in the algorithm space using a virtual edge extension method. This is combined with a nonlinear removal function library dynamically scheduled according to the tool suspension ratio to solve the boundary condition truncation problem of the convolution algorithm at the edge of the microstructure. This achieves accurate compensation for the nonlinear distribution of pressure in the contact area, thereby suppressing the excessive removal of edge material at the physical level and ensuring the conformal accuracy of the geometric features of the microstructure. In the edge-sensitive area, a Peano fractal path with isotropic characteristics is used to replace the conventional grating scanning path. By randomly modulating the motion direction, the coherent superposition condition of the residual texture is destroyed, and the regular mid-frequency ripple error is transformed into high-frequency randomly distributed background noise. This reduces the spectral concentration of the surface morphology, improves the removal efficiency of the subsequent smoothing process, and optimizes the performance indicators of the optical system.

[0023] 3. By utilizing a residual-driven adaptive correction mechanism, online detection feedback is deeply coupled with the process model library. When the real-time processing error exceeds a preset threshold, an asynchronous background calculation thread is triggered to generate optimized parameter pairs and perform a write-back operation to incrementally update the index library. This allows the system to avoid local optima traps caused by measurement noise through data reuse, enabling continuous accumulation of process knowledge and autonomous evolution of control strategies. By establishing a vectorized mapping system from geometric topological features to physical removal models, the coordinate differences of different microstructures in scale and spatial location are shielded, allowing the same set of control logic to adapt to various heterogeneous surface features. This simplifies the system configuration process of precision manufacturing units and enhances the process versatility and operational stability of processing equipment when handling complex topological structures with high aspect ratios. Attached Figure Description

[0024] Figure 1 This is a flowchart of the microstructure conformal polishing method based on white light interference point cloud feedback according to the present invention;

[0025] Figure 2This is a block diagram illustrating the modular architecture and data interaction principle of the microstructure conformal polishing system of the present invention.

[0026] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0028] An adaptive polishing method for fine structures based on point cloud topology awareness includes the following steps:

[0029] Step S1: Obtain white light interference point cloud data of the edge region of the microstructure array on the surface of the workpiece to be processed, and construct a curvature tensor field characterizing the local geometric topological boundary of the microstructure based on the white light interference point cloud data;

[0030] Step S2: Perform dimensionality reduction mapping on the curvature tensor field to extract feature vectors that characterize the spatial morphological properties of the microstructure. The feature vectors have rotation and translation invariance.

[0031] Step S3: Input the feature vector into the preset microstructure edge model association database, and perform an approximate nearest neighbor search in the hash space using the locality sensitive hash algorithm to obtain the matching energy allocation parameters. The energy allocation parameters include the output current adjustment amplitude of the power supply regulation circuit and the removal function kernel parameters used to limit the material removal efficiency of the microstructure unit.

[0032] Step S4: Calculate the power load pulse width required for the power supply regulation circuit to drive the polishing load system based on the energy allocation parameters. The power load pulse width corresponds to the dwell time of the polishing tool in the edge region. Generate the Peano fractal path based on the isotropic constraint conditions of the edge region.

[0033] Step S5: Based on the residence time, the power supply regulation circuit outputs a drive signal with amplitude modulation after output current regulation, which drives the polishing load system to perform material removal along the Peano fractal path. The duration of the drive signal is controlled by the pulse width to achieve electrical energy compensation for nonlinear pressure jumps in the edge region of the microstructure.

[0034] Preferably, step S1 specifically includes: scanning the microstructure array of the workpiece to be processed using an in-situ integrated white light interferometer to obtain a set of three-dimensional coordinate points with a sampling accuracy of not less than 1 nm; calculating the normal vector and principal curvature of each sampling point in the three-dimensional coordinate point set; constructing a curvature tensor field using the principal curvature H to achieve geometric topological characterization of the heterogeneous surface features in the microstructure array; the principal curvature H is determined by the following formula: ,in, The maximum curvature of the sampling point in the principal direction. The minimum curvature of the sampling point in the principal direction.

[0035] Preferably, in step S3, the microstructure edge model association database includes: a key field for storing preset geometric feature vector clusters, which are mapped to multiple hash buckets using a locality-sensitive hashing algorithm; and a value field for storing standard process model parameters that correspond one-to-one with the geometric feature vector clusters. The approximate nearest neighbor retrieval process follows the following mapping rules: ,in, To retrieve the obtained energy allocation parameters, Let f be the feature vector, and let f represent the distance metric function in the hash space.

[0036] Preferably, step S4 specifically includes: performing edge extension for the microstructure topological features in the algorithm space to construct a continuous metasurface to eliminate the boundary condition truncation caused when the polishing load system moves to the edge region; calling the target removal function from the preset nonlinear removal function library according to the tool suspension ratio of the polishing tool; and using the target removal function to correct the dwell time to suppress excessive removal of material in the edge region.

[0037] Preferably, in step S4, the irregular scanning of the Peano fractal path converts the regular mid-frequency ripple error generated in the edge region into high-frequency randomly distributed background noise, thereby reducing the spectral concentration of the surface morphology of the workpiece to be processed and optimizing the modulation transfer function.

[0038] Preferably, the method further includes: step S6, after the material removal action is completed, using white light interference point cloud data to detect processing errors; when the processing error exceeds a preset accuracy threshold, triggering a residual-driven adaptive correction mechanism to generate optimized parameter pairs.

[0039] Preferably, the adaptive correction mechanism performs the following actions through a background asynchronous computing thread: calculates the mapping residual between the processing error and the expected removal amount, and performs incremental iteration on the energy allocation parameters based on the mapping residual; updates the value range in the edge model association database to realize the evolution of process model parameters.

[0040] Preferably, the polishing load system includes an airbag polishing head. In step S5, the pressure distribution of the material removal action follows the Preston equation modified by the energy distribution parameter, and the pressure fluctuation range of the polishing load system is controlled between 0.01 MPa and 0.05 MPa.

[0041] Preferably, in step S2, a topological feature vector extraction algorithm is used to mask the coordinate differences of the microstructure array in absolute space in order to extract feature vectors.

[0042] Preferably, the method is used to process optical components containing heterogeneous microstructure arrays, and to increase the control parameter response frequency of the power supply regulation circuit to above 100Hz through near nearest neighbor search.

[0043] Example 1: In the precision machining of optical components for fused silica microlens arrays, due to the microscale feature size of the microstructure units being on the order of micrometers, the contact state of the polishing head rapidly changes from full-surface contact to local point / line contact as it moves along the edges. To quantify this pressure abrupt change at the microscale, a conformal factor for the microstructure edge is defined. : ,in, The radius of curvature of the local microstructure obtained in step S1 is... For the characteristic lateral dimension of the microstructural unit, when When the preset mutation threshold is exceeded, the system matches the energy allocation parameters using a locality-sensitive hashing algorithm. It will automatically increase the current regulation amplitude of the power supply regulation circuit when When the output current is reduced, the retrieved output current adjustment amplitude is subjected to nonlinear loss correction to prevent over-ablation of the tip of the micro-feature. The polishing load system needs to cope with the challenges of nonlinear pressure distribution and boundary condition truncation caused by abrupt curvature changes in the edge region of the microstructure. The surface of the workpiece to be processed is distributed with periodic microlens units with a sag height of 50μm to 200μm, and there are high gradient geometric feature changes in the edge region. The in-situ white light interferometer integrated into the Z-axis motion unit performs sub-aperture scanning and texture feature-based stitching on the workpiece surface to obtain a set of three-dimensional coordinate points with a sampling accuracy of not less than 1nm. The local topology encoder calculates the normal vector and principal curvature of the sampling points, and uses the formula Construct a curvature tensor field representing the geometric and topological boundaries of the edge region, where H is the principal curvature. The maximum curvature of the sampling point in the principal direction. The minimum curvature of the sampling points along the principal direction is represented by the curvature tensor field, which is then transformed into an eigenvector with rotation and translation invariance through dimensionality reduction mapping. This vector masks the coordinate differences of the microstructure in absolute space, representing only the distribution of its steepness and curvature gradient; the eigenvector... The edge model associated database, used as the query key, stores geometric feature vector clusters mapped to multiple hash buckets using a locality-sensitive hashing algorithm in its key field, and the corresponding standard process model parameters in its value field. The system performs approximate nearest neighbor retrieval within the hash space, following the mapping rules. Obtain matching energy allocation parameters ,in The retrieved energy allocation parameters, f, represent the distance metric function in the hash space. The retrieved energy allocation parameters include the output current regulation amplitude of the power supply regulation circuit and the kernel parameter of the removal function used to limit the material removal efficiency, replacing the real-time numerical inversion of the edge physics field.

[0044] The system performs edge extension in the algorithm space based on energy allocation parameters to construct a continuous metasurface. It calculates the power load pulse width required for the power conditioning circuit to drive the polishing load system, which corresponds to the dwell time of the polishing tool in the edge region. The system generates a Peano fractal path with space-filling characteristics based on the isotropic constraint conditions of the edge region. The power conditioning circuit outputs a drive signal with amplitude modulation adjusted by the output current, which drives the polishing load system containing the airbag polishing head to run along the Peano fractal path. The duration of the drive signal is controlled by the power load pulse width to achieve electrical energy compensation for the nonlinear distribution of pressure in the edge region. The irregular scanning trajectory of the Peano path destroys the coherent superposition conditions of the residual texture of the processing, and transforms the regular mid-frequency ripple error into high-frequency random background noise, so that the collapse error of the edge region is controlled within the preset accuracy threshold range.

[0045] Example 2: This example verifies the engineering effectiveness of the point cloud topology-based adaptive polishing method for microstructures in suppressing edge collapse and eliminating mid-frequency errors by constructing a rigorous experimental system including a multi-dimensional control group and boundary condition stress tests. It also provides measured data support for the rationality of the key process parameter ranges in this invention. The experimental platform is built on a customized five-axis CNC polishing machine, which integrates a white light interferometry unit with a vertical resolution better than 0.1 nm. Its X and Y axis positioning accuracy is calibrated to ±0.5 μm. The test object has a diameter of 100 mm and an edge region with a curvature radius change rate exceeding 0.5 mm. -1To reproduce the non-ideal working conditions of a real industrial environment and verify the system's anti-interference capability, a random micro-vibration noise with a frequency of 50Hz and an amplitude of 0.02mm was actively injected into the machine tool base via a vibrator during the experiment. The processing environment temperature was set within a fluctuation range of 22±1.5℃. The experiment established the baseline logic for parameter decision-making. For the setting of energy allocation parameters, the following engineering trade-off principles were followed: the output current adjustment amplitude of the power supply regulation circuit directly determines the peak value of the material removal rate, while the power load pulse width controls the time domain of energy application. If the current amplitude is too low, it cannot compensate for the contact pressure loss caused by the high curvature of the edge; if the amplitude is too high, it will cause local heat accumulation and subsurface damage. Based on the edge model association database established in the early stage, this experiment sets the effective control range of the output current adjustment amplitude to 0.2A to 2.5A, and the response range of the power load pulse width to 2ms to 50ms. The closed-loop control process of this invention is executed during the experiment: the point cloud of the edge region is obtained with a lateral resolution of 5μm using an in-situ white light interferometer, and the principal curvature H of each sampling point is calculated by a local topology encoder to construct a curvature tensor field.

[0046] The matching energy allocation parameters are retrieved using the locality-sensitive hashing algorithm. The polishing load system was driven to perform material removal along a Peano fractal path. To demonstrate the synergistic effect of the technical features, three levels of control groups were set up in the experiment. Control group one adopted the constant parameters + grating path mode of the existing technology, with a constant polishing pressure of 0.15 MPa and a path spacing of 0.5 mm. The measured data showed that due to the abrupt acceleration change of the grating path at the edge turning point and the overcompensation of the constant pressure in the high curvature region, the material removal in the edge region exhibited a nonlinear surge. Interferometry results showed that the edge collapse value (ERO) was as high as 382.4 nm, and the power spectral density (PSD) curve was at 0.5 mm. -1 Up to 2mm -1 Amplitudes exceeding 15 nm appeared in the spatial frequency range. 2 The periodic characteristic peak at 100 nm indicates residual mid-frequency ripple error on the surface. Control group 2 uses a constant parameter + Peano path mode to isolate the independent contribution of path planning features. Data shows that although the isotropic characteristics of the Peano path reduce the PSD peak to 9.2 nm... 2While improving the mid-frequency error, the edge collapse value remained at 156.7 nm due to the lack of active energy modulation targeting local curvature, failing to meet the requirement of less than 50 nm for high-precision optical components. This result strongly demonstrates that simply changing the path without energy compensation cannot independently solve the edge collapse problem. In contrast, the data from the sample group of this invention showed superior convergence. Under the same environmental disturbance, through dual dynamic modulation of current amplitude and pulse width, the instantaneous energy density at high curvature edges was precisely suppressed. The measured data showed that the edge collapse value was significantly reduced to 34.8 nm, which is about 77.8% better than the control group. At the same time, the PSD curve showed smooth background noise characteristics across the entire frequency band, and the characteristic peaks completely disappeared, indicating that the Peano path and energy modulation mechanism produced a non-obvious synergy. To enhance efficiency, maintain edge geometry, and eliminate machining texture, further verification of the critical significance of the parameter range of this invention was conducted, including an out-of-range boundary pressure test. In boundary test group A, the output current adjustment amplitude was forcibly increased to 3.2A (exceeding the preferred upper limit). Interference detection data showed that the surface roughness (RMS) of the edge region deteriorated sharply from 1.1nm in the sample group of this invention to 5.9nm. Microscopic observation revealed the presence of microcracks caused by instantaneous thermal shock, confirming the necessity of setting the upper limit of the current amplitude. In boundary test group B, the power load pulse width was limited to less than 1ms (below the preferred lower limit). The results showed that the edge correction rate decreased by 85%, and the morphology convergence could not be completed within a reasonable processing cycle, confirming the decisive role of the pulse width lower limit in processing efficiency.

[0047] Example 3: In the offline construction and parameter calibration process of the edge model association database, to eliminate the empirical dependence on the selection of energy allocation parameters in real-time processing, a standardized database filling procedure based on physical simulation and experimental calibration was established. The input of this procedure is defined as covering 0.1mm. -1 Up to 2.0mm -1 A set of standard microstructure templates with varying curvature ranges, covering typical topological features such as steps, grooves, and freeform surface edges, is used. For each standard template, its precise geometric entity is reconstructed using 3D modeling software and imported into a multiphysics simulation engine. In the simulation environment, the edge region is analyzed into independent physical units through discretization meshing. For each unit, a material removal rate model is constructed based on a variant of the Preston equation, where the contact pressure distribution is set as a dynamic variable that is nonlinearly related to the local principal curvature H.

[0048] Based on this, the core parameters are optimized iteratively. For micro-elements with specific curvature tensor characteristics, initial power regulation circuit output parameters including reference current amplitude and pulse duty cycle are set. The simulation engine calculates the theoretical removal function (TIF) generated by the airbag polishing head per unit dwell time under the drive of these parameters, and compares the theoretical removal function with the preset target removal amount. The least squares method is used as the convergence criterion, and the output current regulation amplitude and power load pulse width are automatically adjusted through the gradient descent algorithm until the root mean square error (RMS) of the theoretical removal profile and the target profile converges to within 5nm. This optimal parameter combination for isotropic removal, determined by numerical simulation, is marked as the standard process model parameters under specific geometric features. To establish a fast index between geometric features and optimal parameters, a Local Sensitive Hash (LSH) mapping mechanism based on random projection is introduced. This generates a set of random projection vectors following a Gaussian distribution, which in turn extracts high-dimensional feature vectors from the standard template. Projecting onto a low-dimensional Hamming space, for each projection dimension, if the projection value is greater than zero, it is encoded as 1; otherwise, it is 0. Continuous geometric features are transformed into discrete binary hash codes as keys, as verified by simulation. The corresponding values ​​are written into the hash bucket. To ensure the reliability of the project, a physical sample verification polishing is performed on each hash bucket parameter set before the database is run. The test results show that when the parameters are retrieved from the database for open-loop control, the edge contour error prediction accuracy reaches more than 95%.

[0049] Example 4: In extreme polishing conditions involving high-frequency dynamic load fluctuations, to ensure the long-term stability of the system and the consistency of processing accuracy, a durability aging test and adaptive compensation procedure for the core components of the power supply regulation circuit was constructed. This procedure was set to be executed under a continuous 1000-hour accelerated aging test environment. During this period, the ambient temperature was periodically varied between 10℃ and 50℃ using a temperature-controlled chamber. A step interference signal with a frequency of 1kHz and an amplitude of 20% of the rated current was injected into the load terminal. The system monitored the on-resistance, gate threshold voltage, and ripple coefficient of the output current of the power transistor in the power supply regulation circuit in real time. The measured data showed that… As aging time progresses, the on-resistance of the power transistor exhibits a slow, exponential increase, leading to a cumulative static deviation between the output current regulation amplitude and the preset command value. To address this physical aging phenomenon, the system does not employ open-loop correction but instead activates a Kalman filter-based online parameter identification module. By utilizing real-time acquired voltage-current response data, the system dynamically estimates the current state of the circuit parameters and updates the compensation coefficient of the output current regulation amplitude in real time based on the state estimate. This ensures that even when device aging causes parameter drift, the actual drive energy output to the polishing load system can still accurately follow the preset energy allocation parameters.

[0050] Meanwhile, to address the wear and elastic modulus decay issues of polishing airbags during long-term service, an online calibration strategy based on real-time contact stiffness monitoring was implemented. Utilizing a miniature force sensor integrated within the polishing head, a rapid contact stiffness calibration is performed during the non-working interval of each processing cycle. This drives the polishing head to press into a standard rigid reference surface with a preset small displacement, and records the force-displacement curve. By differentiating this curve, the current equivalent contact stiffness of the airbag is calculated. When the contact stiffness deviates from the initial calibration value by more than 5% due to wear or material aging, the system automatically triggers a reconstruction process for the removal function kernel parameters. This process recalculates the contact area and pressure distribution of the airbag under different compression levels based on the current contact stiffness value, and accordingly corrects the mapping weights used in the local topology encoder to generate the curvature tensor field. Experimental results show that throughout the entire lifespan of the airbag, this online calibration strategy consistently controls the removal depth error of edge polishing within ±3%, avoiding nonlinear degradation of processing quality due to tool wear. This demonstrates that the system possesses adaptive repair and accuracy maintenance capabilities in dynamic, time-varying environments.

[0051] Example 5: In the engineering procedures for ensuring the rapid deployment and stable operation of the adaptive polishing system in different application scenarios, a standardized initial calibration and boundary self-check process for key physical parameters and process control parameters was established. Reference plane calibration was performed on the in-situ white light interferometry unit. A standard flat crystal with a surface shape error less than λ / 20 (λ=632.8nm) was used as the physical reference to calibrate the system error of the interferometer across the entire field of view. Least square fitting was used to remove the tilt and aberration components of the system's optical path, ensuring its accurate measurement of the height information of fine structures. The repeatability is better than 0.5nm. Secondly, the spatial consistency of the removal function of the polishing airbag is calibrated. Under standard working conditions, the airbag is pressed into a textureless planar sample with a constant pressure of 0.15MPa and a fixed-point dwell removal experiment is performed. The three-dimensional morphology of the removal pit is measured by an interferometer, and the half-width at half-maximum and volume removal rate of the removal function are extracted by Gaussian fitting. This calibration process is repeated at different attitude angles of the airbag to construct a correction matrix for the removal function as the attitude changes, thereby eliminating the fluctuation of removal efficiency caused by airbag installation error or uneven wear.

[0052] Finally, a self-test is performed on the dynamic response boundary of the power supply regulation circuit. Under no-load conditions, a series of test signals with frequencies ranging from 1Hz to 1kHz and amplitudes increasing from 0.1A to 3.0A are input to the power supply regulation circuit. By monitoring the waveform distortion rate and phase lag of the output current, the maximum effective bandwidth and upper limit of the current amplitude within the linear response region are determined. When the measured response parameters exceed the preset safety threshold, the system automatically triggers an alarm and limits the processing speed to prevent edge overshoot or under-removal caused by insufficient dynamic performance of the circuit. To ensure high-frequency response accuracy, a pneumatic-electromagnetic dual-stage hybrid actuation architecture is adopted, and a frequency domain response frequency division calibration procedure is executed. The architecture defines the physical division of labor between the low-frequency flexible component and the high-frequency actuation component: the airbag polishing head acts as the low-frequency flexible medium, providing static pre-pressure and adapting to surface shape changes below 5Hz; the voice coil motor (VCM) at the end of the series Z-axis drive chain acts as the high-frequency actuator, responding to the high-frequency current command output by the power supply regulation circuit and executing micro-amplitude pressure modulation above 100Hz. The procedure is as follows: Under static contact conditions, input a sinusoidal sweep signal with a frequency range of 1Hz to 500Hz and an amplitude of 10% of the rated current to the voice coil motor driver; use a high-frequency force sensor at the flange of the integrated polishing head to sample at a frequency of not less than 5kHz and collect the output pressure waveform in real time; construct the Bode plot of the system's amplitude-frequency characteristic and phase-frequency characteristic curves, and define the effective control bandwidth of the closed-loop system as the frequency point where the amplitude attenuation is -3dB and the phase lag does not exceed 45°; based on the measured Bode plot, determine the differential gain and feedforward coefficient of the PID controller in the power supply regulation circuit. Experimental data shows that under the condition of a constant airbag inflation pressure of 0.1MPa, the response delay of the two-stage architecture to the force control command of 100Hz to 150Hz is less than 2ms. It is physically feasible to use the high dynamic characteristics of the voice coil motor to compensate for the instantaneous nonlinear pressure at the edge. This series of pre-calibration and self-testing procedures provides a deterministic physical benchmark and control boundary for subsequent adaptive edge conformal polishing, ensuring the reproducibility of the process scheme under different equipment and environmental conditions.

[0053] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for topology-aware adaptive polishing of microstructures based on point cloud, characterized in that, Includes the following steps: Step S1: Obtain white light interference point cloud data of the edge region of the microstructure array on the surface of the workpiece to be processed, and construct a curvature tensor field characterizing the local geometric topological boundary of the microstructure based on the white light interference point cloud data. Also, use an in-situ integrated white light interferometer to scan the microstructure array of the workpiece to be processed to obtain a set of three-dimensional coordinate points with a sampling accuracy of not less than 1 nm. The normal vector and the principal curvature of each sampling point in the three-dimensional coordinate point set are calculated, a curvature tensor field is constructed by using the principal curvature H, and geometric topological characterization of heterogeneous surface features in the microstructure array is realized; the principal curvature H is determined by the following formula: wherein, is the maximum curvature of the sampling point in the principal direction, is the minimum curvature of the sampling point in the principal direction; Step S2: Perform dimensionality reduction mapping on the curvature tensor field to extract feature vectors that characterize the spatial morphological properties of the microstructure. The feature vectors have rotation and translation invariance. Step S3: Input the feature vectors into a preset microstructure edge model association database, and perform an approximate nearest neighbor search in the hash space using the Locality Sensitive Hashing (LSH) algorithm to obtain matching energy allocation parameters. These energy allocation parameters include the output current adjustment amplitude of the power supply regulation circuit and the kernel parameters of the removal function used to limit the material removal efficiency of the microstructure unit. The microstructure edge model association database includes: a key field for storing preset geometric feature vector clusters, which are mapped to multiple hash buckets using the LSH algorithm; and a value field for storing standard process model parameters that correspond one-to-one with the geometric feature vector clusters. The approximate nearest neighbor search process follows the following mapping rules: ,in, To retrieve the obtained energy allocation parameters, Let f be the feature vector, and let f represent the distance metric function in the hash space; Step S4: Calculate the power load pulse width required for the power supply regulation circuit to drive the polishing load system based on the energy allocation parameters. The power load pulse width corresponds to the dwell time of the polishing tool in the edge region. Generate the Peano fractal path based on the isotropic constraint conditions of the edge region. Step S5: Based on the residence time, the power supply regulation circuit outputs a drive signal with amplitude modulation after output current regulation, which drives the polishing load system to perform material removal along the Peano fractal path. The duration of the drive signal is controlled by the pulse width to achieve electrical energy compensation for nonlinear pressure jumps in the edge region of the microstructure.

2. The method of claim 1, wherein, Step S4 specifically includes: performing edge extension targeting the microstructure topological features within the algorithm space to construct a continuous metasurface to eliminate boundary condition truncation caused when the polishing load system moves to the edge region; calling the target removal function from the preset nonlinear removal function library according to the tool suspension ratio of the polishing tool; and using the target removal function to correct the dwell time to suppress excessive removal of material in the edge region. 3.The microstructure adaptive polishing method based on point cloud topology perception according to claim 1, wherein, In step S4, the irregular scanning of the Peano fractal path transforms the regular mid-frequency ripple error generated in the edge region into high-frequency randomly distributed background noise, thereby reducing the spectral concentration of the surface morphology of the workpiece to be processed and optimizing the modulation transfer function.

4. The method of claim 1, wherein, It also includes: step S6, after the material removal action is completed, using white light interference point cloud data to detect processing errors; when the processing error exceeds the preset accuracy threshold, triggering the residual-driven adaptive correction mechanism to generate optimized parameter pairs.

5. The method of claim 4, wherein, The adaptive correction mechanism performs the following actions through a background asynchronous computing thread: calculates the mapping residual between the processing error and the expected removal amount, and performs incremental iterations on the energy allocation parameters based on the mapping residual; Update the value range in the edge model association database to realize the evolution of process model parameters.

6. The method of claim 1, wherein, The polishing load system includes an airbag polishing head. In step S5, the pressure distribution of the material removal action follows the Preston equation modified by the energy distribution parameter, and the pressure fluctuation range of the polishing load system is controlled between 0.01 MPa and 0.05 MPa.

7. The method of claim 1, wherein, In step S2, a topological feature vector extraction algorithm is used to mask the coordinate differences of the microstructure array in absolute space in order to extract feature vectors.

Images