Fine structure conformal polishing method based on in-situ white light interference point cloud feedback

Abstract

The invention relates to the technical field of ultra-precision polishing, in particular to a fine structure conformal polishing method based on in-situ white light interference point cloud feedback, which comprises the following steps: S1, splicing and acquiring sub-apertures of global morphology; s2, automatic identification and virtual continuation of an edge sensitive area; s3, carrying out residence time inversion based on a suspension ratio; s4, generating a Peano random modulation path; and S5, closed-loop iteration verification is carried out. According to the method, boundary truncation errors are eliminated through virtual edge continuation, edge pressure mutation is accurately compensated in combination with a suspension ratio-based variable-pressure TIF library, so that the edge collapse amount is controlled within 50nm, and the geometric accuracy of the microstructure is remarkably improved; a Piano fractal path is adopted to replace a traditional grating path, edge residual errors are converted into high-frequency noise, the high-frequency noise can be removed through subsequent full-area smoothing within extremely short time, and it is ensured that a surface modulation transfer function meets the requirement.

Description

Technical Field

[0001] This invention relates to the field of ultra-precision polishing technology, specifically to a microstructure conformal polishing method based on in-situ white light interference point cloud feedback. Background Technology

[0002] Currently, ultra-precision polishing technology faces a triple dilemma of "efficiency-precision-damage" in the processing of microstructures: traditional contact polishing is highly efficient (material removal rate can reach 5μm / min), but it is prone to edge collapse due to concentrated edge pressure (typical error 80-150nm); non-contact polishing can avoid mechanical damage, but it suffers from low processing efficiency (removal rate only 0.1-0.5μm / min) and subsurface damage layer (depth > 100nm); while existing adaptive control algorithms mostly focus on planar areas and have insufficient precision control over the edges of microstructures (curvature radius < 5μm), making it difficult to simultaneously achieve the required surface roughness (Ra) and shape accuracy (PV value).

[0003] To address the technical bottlenecks in the polishing process of microstructured parts, such as edge collapse, boundary truncation error, and mid-frequency residue, we propose a conformal polishing method for microstructures based on in-situ white light interference point cloud feedback. Summary of the Invention

[0004] The purpose of this invention is to provide a microstructure conformal polishing method based on in-situ white light interference point cloud feedback, so as to solve the technical bottlenecks in the polishing process of microstructure parts, such as edge collapse, boundary truncation error, and intermediate frequency residue, as mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a microstructure conformal polishing method based on in-situ white light interferometric point cloud feedback, comprising the following steps: S1: acquisition of sub-aperture stitching of global topography: control the linkage between the C-axis rotation unit and the Z-axis motion unit, use a white light interferometer to scan the sub-aperture of the microstructure workpiece surface, and use a feature constraint stitching algorithm based on the inherent periodic texture of the microstructure to stitch the sub-aperture scanning data into global three-dimensional point cloud data P; S2: Automatic identification and virtual extension of edge-sensitive areas: Curvature analysis is performed on the global 3D point cloud data P, and the curvature change points are extracted as structural edges. The regions on both sides of the edge are defined as edge-sensitive areas. Along the normal of the structural edge, based on the tangent and curvature of the adjacent surface, a virtual point cloud P is generated by polynomial fitting and filled to the outside of the original point cloud data to construct a continuous mathematical metasurface. S3: Dwell Time Inversion Based on Suspension Ratio: For each point within the edge-sensitive zone, calculate the suspension ratio α of the tool relative to the physical edge; call the pre-set pressure removal function library R, ​​which is calibrated through offline finite element analysis and single-point experiments to describe the nonlinear pressure distribution under different suspension ratios α; construct the matrix equation H=R(α)·D, and use the Tikhonov regularization algorithm with non-negativity constraints and edge smoothing constraints to solve for the dwell time D; S4: Generate Peano random modulation path: Map the obtained dwell time D to machine tool motion command. Standard machining path is used in non-edge areas and Peano fractal path is used in edge sensitive areas. Dynamically adjust the feed rate F so that the actual dwell time is close to the calculated value D(x,y), and the feed rate F reaches the maximum value when crossing edge singularities to achieve zero removal of crossing. S5: Closed-loop iterative verification: After polishing, the residual error in the edge-sensitive area is detected again by white light interferometer. If the edge collapse exceeds the preset threshold, local compensation processing code is automatically generated, and the process returns to steps S3-S4 for secondary correction until the residual error meets the requirements.

[0006] Preferably, in step S2, the polynomial fitting uses a cubic or quintic polynomial, and the extension length of the virtual point cloud P{virtual} is 1-2 times the tool radius.

[0007] Preferably, in step S3, the suspension ratio α is the ratio of the area of ​​the tool extending beyond the physical edge to the total contact area of ​​the tool, and the value range is 0-1.

[0008] Preferably, in step S3, the regularization parameter λ of the Tikhonov regularization algorithm takes a value in the range of 10. 6 -10 4 .

[0009] Preferably, in step S4, the fractal dimension of the Peano fractal path is 2, and the path step length is dynamically adjusted according to the dwell time D(x,y), with an adjustment range of 0.1-1μm.

[0010] Preferably, in step S5, the preset threshold is 30-50nm.

[0011] Preferably, in step S1, the positioning accuracy of the C-axis rotation unit is ±1 sec, the scanning field of view of the white light interferometer is 50-200 μm, and the scanning resolution is ≤0.1 nm.

[0012] Preferably, in step S3, the transformer removal function library R(α) contains at least 10 sets of nonlinear removal models corresponding to different suspension ratios α, and the sampling interval of α is ≤0.1.

[0013] Preferably, the microstructured workpiece includes grating elements, microlens arrays, and high aspect ratio microgroove parts, with microstructure feature dimensions of 1-100 μm.

[0014] Preferably, the polishing machine includes a polishing machine housing, a C-axis rotation unit, a Z-axis motion unit, and an edge-sensitive area. The C-axis rotation unit is installed inside the polishing machine housing, and the Z-axis motion units are installed on both sides of the C-axis rotation unit on the surface of the polishing machine housing. The edge-sensitive area is installed on the inner wall of the polishing machine housing.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention eliminates boundary truncation errors by virtual edge extension and, combined with a variable TIF library based on suspension ratio, accurately compensates for sudden changes in edge pressure, keeping the edge collapse amount within 50nm and significantly improving the geometric accuracy of the microstructure.

[0016] This invention uses a Peano fractal path to replace the traditional grating path, converting residual edge errors into high-frequency noise, which can be removed by subsequent full-area smoothing in a very short time, ensuring that the surface modulation transfer function meets the requirements.

[0017] This invention integrates in-situ detection of white light interferometry with algorithm feedback, and can achieve adaptive polishing of complex microstructures without the need for physically adding "sacrificial blocks" (which cannot be added inside the microstructure), thus exhibiting strong adaptability.

[0018] The partitioned path planning of this invention not only ensures the processing efficiency of non-edge areas, but also focuses on solving the accuracy problem of edge areas. Compared with traditional methods, the processing efficiency is improved by more than 20%, and the edge processing accuracy is improved by an order of magnitude.

[0019] This invention does not require modification of the core hardware of existing CCOS machine tools; it can be achieved simply by upgrading the control system algorithm. This makes it easy to promote and apply, and it is suitable for processing various micro-structured parts such as grating elements, microlens arrays, and high aspect ratio microgrooves. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the polishing machine of the present invention; Figure 3This is a top view schematic diagram of the polishing machine of the present invention; Figure 4 This is a front view schematic diagram of the polishing machine of the present invention.

[0022] In the diagram: 1. Polishing machine housing; 2. C-axis rotation unit; 3. Z-axis motion unit; 4. Edge sensitive area. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1-4 This invention provides an embodiment of a microstructure conformal polishing method based on in-situ white light interferometry point cloud feedback, comprising the following steps: S1: Obtaining sub-aperture stitching of the global topography: Controlling the C-axis rotation unit 2 and the Z-axis motion unit 3 to perform sub-aperture scanning on the surface of the microstructure workpiece using a white light interferometer (WLI) (scanning step size 50-100μm, overlap rate ≥30%), employing a feature constraint stitching algorithm based on the inherent periodic texture of the microstructure (registration accuracy ≤0.5 pixels), eliminating stitching accumulation errors through multi-scale iterative optimization, and stitching the sub-aperture scanning data into global three-dimensional point cloud data P(x,y,z) (point cloud density ≥10). 6 points / mm 2 ); S2: Automatic identification and virtual extension of edge-sensitive region 4: After Gaussian filtering (σ=0.8-1.2μm) of the global 3D point cloud data P(x,y,z), curvature analysis is performed, and the curvature abrupt change point is extracted using the 8-neighborhood gradient method (curvature threshold 10). -3 The area ( / mm) is defined as the edge of the structure, with a 50-200μm region on each side of the edge as the edge sensitive zone 4 (ESZ). Along the normal direction of the structure edge outwards, based on the tangent and curvature of adjacent surfaces, a virtual point cloud Pvirtual is generated by fitting a cubic polynomial (planar edge) or a quintic polynomial (curved surface edge). This virtual point cloud is then filled outside the original point cloud data to construct a continuous mathematical metasurface (fitting residual ≤5nm). Curvature analysis is used to automatically identify microstructure edges (curvature abrupt change points, threshold 10). -3 / mm), defining the 50-200μm area on both sides of the edge as the edge sensitive zone (ESZ). Along the edge normal outward, a cubic polynomial (planar edge) or a fifth polynomial (curved edge) is used to fit and generate a virtual point cloud, with an extension length of 1-2 times the tool radius, to construct a continuous mathematical metasurface (fitting residual ≤5nm), eliminate boundary truncation error, and ensure the geometric fit between the polishing tool and the edge contour; S3: Dwell Time Inversion Based on Suspension Ratio: For each point within the edge-sensitive zone 4 (ESZ), the Monte Carlo method is used to calculate the suspension ratio α of the tool relative to the physical edge (number of sampling points ≥ 1000); the pre-set pressure removal function library R(α) is called, and the pressure removal function library R(α) is calibrated through offline finite element analysis (FEA) (mesh accuracy 1μm) and single-point experiments (error ≤ 3%) to describe the nonlinear pressure distribution under different suspension ratios α (pressure range 0.1-2N); the matrix equation H=R(α)·D is constructed. The dwell time D(x,y) was solved using a Tikhonov regularization algorithm with nonnegativity constraints and edge smoothing constraints (smoothing factor 0.01-0.1) (calculation error ≤ 2ms). The tool overhang ratio α (the ratio of the area of ​​the tool extending beyond the physical edge to the total contact area, 0-1) within the ESZ was calculated. A transformer removal function library R(α) calibrated through offline finite element analysis (FEA) and experiments (containing at least 10 nonlinear removal models, α sampling interval ≤ 0.1) was called to dynamically adjust the pressure distribution (0.1-2N). This was combined with the Tikhonov regularization algorithm (regularization parameter λ=10). 6 -10 4 Solve for the dwell time D(x,y) to ensure a smooth pressure transition when the tool contacts the edge, avoiding edge collapse caused by local overpressure; S4: Generate Peano random modulation path: Map the obtained dwell time D(x,y) to machine tool motion commands (pulse equivalent 0.1μm). Standard grating processing path (row spacing 5-10μm) is used in non-edge areas, and Peano fractal path (corner radius ≥5μm) is used in edge sensitive area 4 (ESZ). The feed speed F is dynamically adjusted based on PID feedback control (adjustment period 1ms) so that the actual dwell time is close to the calculated value D(x,y) (deviation ≤5%). When crossing the edge singularity, the feed speed F reaches its maximum value (20-30mm / s) to achieve zero removal crossing (removal amount ≤0.1nm). A Peano path with fractal dimension of 2 is used in ESZ. The path step size is dynamically adjusted according to the dwell time (0.1-1μm), which reduces the edge texture error by 30% compared with the traditional grating path. By controlling the feed speed through PID feedback (adjustment period 1ms), the speed is increased to 20-30mm / s at the edge singularity to achieve "zero removal jump" (removal amount ≤0.1nm), ensuring that the trajectory closely fits the microstructure contour; S5: Closed-loop iterative verification: After polishing, the residual error of the edge-sensitive area 4 (ESZ) is detected again using a white light interferometer (WLI) (sampling frequency 1kHz). If the edge collapse amount (PV value) exceeds the preset threshold (30-50nm), local compensation processing code is automatically generated (compensation coefficient 0.8-1.2), and the process returns to steps S3-S4 for secondary correction (iteration count ≤ 3 times) until the residual error meets the requirements (PV ≤ 30nm, Ra ≤ 1nm). After polishing, the residual error of ESZ is detected again using a white light interferometer (WLI). If the edge collapse amount (PV value) exceeds the preset threshold (30-50nm), the residual error is detected again. If the value exceeds the 30-50nm threshold (30nm for optical components and 50nm for MEMS components), the device automatically generates compensation code and performs secondary correction until the shape accuracy (PV≤30nm) and surface quality (Ra≤1nm) meet the standards, achieving high-precision fitting of the polishing effect with the design morphology. The device achieves synchronization of sub-aperture scanning, path execution and in-situ detection through the coordinated control of the C-axis rotation unit 2 (positioning accuracy ±1sec), Z-axis motion unit (positioning accuracy ±0.1μm) and edge sensitive area detection module (response time ≤10ms), providing hardware guarantee for structural fitting.

[0025] This device solves the technical problems of collapse deformation caused by sudden pressure changes in the edge region during traditional polishing (reducing edge error), ill-conditioned problems in dwell time solution (reducing the number of conditions by two orders of magnitude), and surface quality fluctuations caused by discontinuous path planning (increasing surface roughness Ra to 0.8 nm) by using virtual edge extension technology and the coordinated control of the pressure change removal model.

[0026] Furthermore, in step S2, the polynomial fitting uses a cubic polynomial (planar structure) or a fifth-order polynomial (curved surface structure), and the extension length of the virtual point cloud Pvirtual is 1-2 times the tool radius. This parameter configuration ensures the continuity of the first derivative between the virtual surface and the real surface, effectively eliminating overcutting caused by boundary effects. The combination of virtual extension and variable pressure compensation controls the edge collapse within 50nm, improving the conformal accuracy of microstructure feature sizes (1-100μm) by an order of magnitude.

[0027] Furthermore, in step S3, the suspension ratio α is the ratio of the area of ​​the tool extending beyond the physical edge to the total contact area of ​​the tool, with a value ranging from 0 to 1. When α=0, it indicates that the tool is completely inside the workpiece; when α=1, it indicates that the tool is completely suspended. This parameter directly determines the dynamic adjustment range of the pressure correction coefficient. The Peano fractal path converts the mid-frequency error into high-frequency noise, which can be removed by smoothing in a short time, ensuring that the surface modulation transfer function meets the requirements of optical / microelectromechanical systems.

[0028] Furthermore, in step S3, the regularization parameter λ of the Tikhonov regularization algorithm takes values ​​ranging from 10. 6 -104 By optimizing the λ value using the L-curve method, the dwell time calculation error can be minimized (≤2%) while ensuring the stability of the solution. This method is particularly suitable for ultra-precision machining of hard and brittle materials (sapphire, silicon wafers) such as grating elements (linewidth 1-50μm), microlens arrays (diameter 10-100μm), and high aspect ratio microtrenches (depth-to-width ratio ≥5:1).

[0029] Furthermore, in step S4, the fractal dimension of the Peano fractal path is 2, and the path step size is dynamically adjusted according to the dwell time D(x,y), with an adjustment range of 0.1-1μm. The fractal path can enable the processing trajectory to form a uniform coverage in the same area, reducing edge texture error by 30% compared to the traditional grating path.

[0030] Furthermore, in step S5, the preset threshold is 30-50nm. This threshold is set according to the application scenario of the workpiece; for optical components, it is usually 30nm, while for microelectromechanical system components, it can be relaxed to 50nm.

[0031] Furthermore, in step S1, the positioning accuracy of the C-axis rotation unit 2 is ±1 sec, and the repeatability is ±0.5 sec; the scanning field of view (FOV) of the white light interferometer (WLI) is 50-200 μm, and the scanning resolution is ≤0.1 nm. High-resolution scanning ensures accurate reconstruction of microstructure features (below 1 μm).

[0032] Furthermore, in step S3, the transformer removal function library R(α) contains at least 10 sets of nonlinear removal models corresponding to different suspension ratios α, with a sampling interval of α ≤ 0.1. Through interpolation algorithms, α can be continuously invoked within the range of 0-1, and the model prediction error is ≤ 5%.

[0033] Furthermore, the microstructured workpieces include grating elements (linewidth 1-50 μm), microlens arrays (diameter 10-100 μm), and high aspect ratio microgroove parts (depth-to-width ratio ≥ 5:1), with microstructure feature dimensions of 1-100 μm. This method is particularly suitable for the ultra-precision machining of hard and brittle materials (such as sapphire and silicon wafers).

[0034] Furthermore, the system includes a polishing machine housing 1, a C-axis rotation unit 2 (maximum speed 300 rpm), a Z-axis motion unit 3 (positioning accuracy ±0.1 μm), and an edge-sensitive area 4. The C-axis rotation unit 2 (capable of supporting workpiece weight ≤5 kg) is installed inside the polishing machine housing 1. The Z-axis motion units 3 (stroke 100 mm) are installed on both sides of the C-axis rotation unit 2 on the surface of the polishing machine housing 1. The edge-sensitive area 4 (detection response time ≤10 ms) is installed on the inner wall of the polishing machine housing 1. All units are synchronously controlled via an industrial Ethernet network (communication delay ≤1 ms).

[0035] Working principle: Step S1: Sub-aperture stitching acquisition of global topography. The X-axis and C-axis are linked. The white light interferometer performs sub-aperture scanning with a step size of 50μm, and a total of 36 sub-aperture data are collected. The periodic arrangement texture of the microlens array is used as the stitching feature points. The feature constraint stitching algorithm is used to obtain global three-dimensional point cloud data P(x,y,z) with a stitching accuracy of 0.1 pixels.

[0036] Step S2: Automatic identification and virtual extension of edge sensitive area (4) Perform curvature analysis on P(x,y,z), extract the curvature change point of the microlens edge, and define the area of ​​5mm (1 times the tool radius) on both sides of the edge as ESZ; along the edge normal outward, generate virtual point cloud P_{virtual} by cubic polynomial fitting, extend the length by 5mm, and construct a continuous metasurface.

[0037] Step S3: Dwell time inversion based on suspension ratio Calculate the suspension ratio α at each point within the ESZ (sampling interval 0.05); Call the transformer removal function library R(α) (which contains 20 sets of calibration data); The target removal amount H is set (5μm for edge region target removal), and the Tikhonov regularization algorithm (λ=5×10) is used. -5 Solve for the dwell time D(x,y) to obtain the dwell time distribution from 0 to 200 ms.

[0038] Step S4: Generate Peano random modulation path Non-edge areas: use grating path, feed speed 10mm / s; ESZ region: Generates a Peano path with a fractal dimension of 2, a step size of 0.5μm, and dynamically adjusts the feed rate (5-20mm / s) according to the dwell time. At the edge singularity: the feed rate is increased to 25mm / s, achieving zero removal jump.

[0039] Step S5: Closed-loop iterative verification After the first polishing, the WLI detection showed that the edge collapse amount PV in the ESZ region was 48nm (exceeding the threshold of 40nm). The control system automatically generated local compensation code and returned to steps S3-S4 for secondary correction (compensation dwell time 5-15ms). After the second polishing, the edge collapse amount PV was 32nm, the surface roughness Ra was 0.008μm, and there was no intermediate frequency error, meeting the target requirements. The above is the complete working principle of this invention.

[0040] 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 invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback, characterized in that: Includes the following steps: S1: Obtaining sub-aperture stitching of global topography: Control the C-axis rotation unit (2) and Z-axis motion unit (3) to perform sub-aperture scanning on the surface of the microstructure workpiece using a white light interferometer (WLI). Use a feature constraint stitching algorithm based on the inherent periodic texture of the microstructure to stitch the sub-aperture scanning data into global three-dimensional point cloud data P(x,y,z). S2: Automatic identification and virtual extension of edge sensitive area (4): Curvature analysis is performed on global three-dimensional point cloud data P(x,y,z), and curvature change points are extracted as structural edges. The regions on both sides of the edge are defined as edge sensitive areas (4) (ESZ). Along the normal of the structural edge, based on the tangent and curvature of the adjacent surface, a virtual point cloud P_{virtual} is generated by polynomial fitting and filled to the outside of the original point cloud data to construct a continuous mathematical metasurface. S3: Dwell time inversion based on suspension ratio: For each point in the edge-sensitive zone (4) (ESZ), calculate the suspension ratio α of the tool relative to the physical edge; call the preset pressure removal function library R(α), which is calibrated by offline finite element analysis (FEA) and single-point experiments to describe the nonlinear pressure distribution under different suspension ratios α; construct the matrix equation H=R(α)·D, and use the Tikhonov regularization algorithm with non-negative constraints and edge smoothing constraints to solve the dwell time D(x,y). S4: Generate Peano random modulation path: Map the obtained dwell time D(x,y) to machine tool motion command. Standard machining path is used in non-edge areas, and Peano fractal path is used in edge sensitive area (4) (ESZ). Dynamically adjust the feed rate F so that the actual dwell time is close to the calculated value D(x,y), and the feed rate F reaches the maximum value when crossing edge singularity to achieve zero removal of crossing. S5: Closed-loop iterative verification: After polishing, the residual error of the edge sensitive area (4) (ESZ) is detected again by white light interferometer (WLI). If the edge collapse amount (PV value) exceeds the preset threshold, the local compensation processing code is automatically generated and the process returns to steps S3-S4 for secondary correction until the residual error meets the requirements.

2. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S2, the polynomial fitting uses a cubic or quintic polynomial, and the extension length of the virtual point cloud P{virtual} is 1-2 times the tool radius.

3. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S3, the suspension ratio α is the ratio of the area of ​​the tool extending beyond the physical edge to the total contact area of ​​the tool, and the value ranges from 0 to 1.

4. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S3, the regularization parameter λ of the Tikhonov regularization algorithm takes values ​​ranging from 10. 6 -10 4 .

5. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S4, the fractal dimension of the Peano fractal path is 2, and the path step size is dynamically adjusted according to the dwell time D(x,y), with an adjustment range of 0.1-1μm.

6. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S5, the preset threshold is 30-50nm.

7. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S1, the positioning accuracy of the C-axis rotation unit (2) is ±1 sec, the scanning field of the white light interferometer is 50-200 μm, and the scanning resolution is ≤0.1 nm.

8. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: In step S3, the transformer removal function library R(α) contains at least 10 sets of nonlinear removal models corresponding to different suspension ratios α, and the sampling interval of α is ≤0.

1.

9. The method for conformal polishing of microstructures based on in-situ white light interference point cloud feedback according to claim 1, characterized in that: The microstructured workpieces include grating elements, microlens arrays, and high aspect ratio microgroove parts, with microstructure feature dimensions ranging from 1 to 100 μm.

10. A polishing machine, comprising a microstructure conformal polishing method based on in-situ white light interference point cloud feedback according to any one of claims 1-9, characterized in that: The polishing machine includes a polishing machine housing (1), a C-axis rotation unit (2), a Z-axis motion unit (3), and an edge-sensitive area (4). The C-axis rotation unit (2) is installed inside the polishing machine housing (1). The Z-axis motion unit (3) is installed on both sides of the C-axis rotation unit (2) on the surface of the polishing machine housing (1). The edge-sensitive area (4) is installed on the inner wall of the polishing machine housing (1).

Images