Crystal weight configuration method and slotted weight block

Abstract

The application is suitable for the field of crystal polishing, and discloses a crystal weight block configuration method and a slotted weight block. The method comprises the following steps: obtaining the unloaded global surface profile data of a crystal to be processed, and extracting surface profile features from the unloaded global surface profile data; dividing the surface of the crystal to be processed into a plurality of sub-zones according to the surface profile features; determining the height value of the global lowest point from the unloaded global surface profile data as a polishing reference height, and determining the expected material removal amount of each sub-zone according to the local high point of each sub-zone and the polishing reference height; inputting the expected material removal amount of each sub-zone into a pre-trained correlation prediction model to obtain the required pressure value of each sub-zone; and determining the slotted weight block corresponding to each sub-zone according to the required pressure value of each sub-zone, and forming a weight block array configuration scheme. The method can output high-precision sub-zone pressure requirements and directly map them into a weight configuration scheme of the slotted weight block.

Description

Technical Field

[0001] This invention relates to the field of crystal polishing, and more particularly to a method for configuring the weight of a crystal counterweight and a slotted counterweight. Background Technology

[0002] Large-size crystals (such as laser crystals and optical crystals) are key components of high-precision optical systems, widely used in high-power lasers, space optical systems, and other fields, serving as core components such as gain media, windows, and prisms. Their surface shape accuracy (including flatness and peak-to-valley value PV) directly affects the system's wavefront distortion, beam quality, and energy efficiency, requiring extremely stringent precision. In inertial confinement fusion devices, the surface shape PV of large-aperture KDP crystals needs to be better than λ / 20 (approximately 30 nm), the PV of gravitational wave detector mirrors needs to be less than 1 nm, and space optical silicon carbide elements need to achieve a PV of 60 nm and a roughness of less than 2 nm RMS. This level of nanometer and even sub-nanometer precision far exceeds traditional processing capabilities, necessitating precision polishing processes such as magnetorheological polishing and ion beam polishing to achieve atomic-level smoothness, correct multi-scale errors, adapt to the processing of brittle and heat-sensitive crystals, and improve system stability and reliability under extreme environments. Precision polishing technology has become an indispensable core element in ensuring the performance of high-precision optical systems.

[0003] Traditional polishing processes have significant shortcomings in the pressure application and measurement stages. For example, pressure application often uses a monolithic, fixed pressure structure, which cannot differentiate and quantify the loading of high points on the crystal surface. This results in uneven pressure distribution in the contact area between the workpiece and the polishing pad during polishing, and random and uncontrollable deformation of the flexible contact, making it difficult to achieve precise surface correction. At the same time, measurement often relies on offline optical or contact equipment, with pressure application and measurement being separate or interfering with each other. It is impossible to obtain accurate deformation data of the pressure area in situ and in real time during processing, and it is difficult to capture the true contact morphology of the workpiece under loading conditions. Consequently, it is impossible to effectively separate the elastic deformation caused by pressure from the actual material removal from the final surface shape. This makes it difficult to clarify the process mechanism, and the material removal rate cannot be accurately correlated with pressure. Ultimately, this results in poor material removal uniformity, difficulty in predicting and guaranteeing surface accuracy, and a high dependence on experience for process parameters, leading to low processing convergence efficiency and high trial-and-error costs.

[0004] Therefore, existing technologies need to be improved and developed. Summary of the Invention

[0005] The first objective of this invention is to provide a method for configuring the weight of a crystal counterweight, which aims to solve the technical problem that the material removal rate cannot be accurately correlated with pressure in existing processing.

[0006] To achieve the above objectives, the solution provided by the present invention is as follows: A method for configuring the weight of a crystal counterweight includes: acquiring unloaded global surface shape data of the crystal to be processed, and extracting surface features from the unloaded global surface shape data; dividing the surface of the crystal to be processed into several partitions according to the surface features; determining the height value of the global lowest point from the unloaded global surface shape data as a polishing reference height, and determining the expected material removal amount of each partition based on the local high point of each partition and the polishing reference height; inputting the expected material removal amount of each partition into a pre-trained correlation prediction model to obtain the pressure value required for each partition; determining the slotted counterweight corresponding to each partition based on the pressure value required for each partition, and forming a counterweight array configuration scheme.

[0007] Preferably, the step of acquiring the unloaded global surface shape data of the crystal to be processed and extracting surface shape features from the unloaded global surface shape data includes: measuring the entire surface of the crystal to be processed using an interferometer to obtain the unloaded global surface shape data of the crystal to be processed and its height data matrix; generating a PV value cloud map based on the height data matrix; and extracting surface shape features from the PV value cloud map, wherein the surface shape features include the position, amplitude, and distribution range of local high points.

[0008] Preferably, the step of dividing the surface of the crystal to be processed into several partitions according to the surface features includes: extracting all local high points of the crystal to be processed from the surface features; dividing the area where the local high points with amplitudes exceeding a preset amplitude threshold are located into independent partitions, and dividing the remaining areas according to a preset grid to obtain several partitions, wherein the total number of partitions does not exceed a preset upper limit.

[0009] Preferably, the step of determining the height value of the global lowest point from the unloaded global surface data as the polishing reference height, and determining the expected material removal amount for each partition based on the local high point of each partition and the polishing reference height, includes: determining the height value of the global lowest point from the unloaded global surface data as the polishing reference height; extracting the highest height value of the local high point within each partition; and for each partition, using the difference between its highest height value and the polishing reference height as the expected material removal amount corresponding to the partition.

[0010] Preferably, the desired material removal amount for each partition is input into a pre-trained correlation prediction model to obtain the required pressure value for each partition. Prior to this, the method includes: acquiring unloaded training surface data of the training crystal and various pressure distribution arrays adapted to the training crystal; for each pressure distribution array, under the loaded state of the slotted counterweight, a laser displacement sensor passes through the slot on the slotted counterweight to perform in-situ measurements on the surface of the training crystal below each partition, obtaining the pre-processing loaded surface data under the loaded state; after the training crystal polishing process is completed, in-situ measurements are performed again under the condition that the slotted counterweight loading state remains unchanged, obtaining the post-processing loaded surface of the training crystal under the loaded state. Shape data; acquire the global surface shape data of the training crystal after processing without load, and compare it with the unloaded training surface shape data. Based on the comparison results, determine whether the training crystal has undergone permanent deformation. When it is determined that the training crystal has not undergone permanent deformation, calculate the elastic deformation and material removal amount based on the unloaded training surface shape data, the pre-processing loaded surface shape data, and the post-processing loaded surface shape data, and form training data pairs. The training data pairs are associated with the pressure, elastic deformation, and material removal amount of the slotted counterweight. The weight combination of the slotted counterweight is changed multiple times to accumulate multiple sets of training data pairs. A machine learning algorithm is used to train an associated prediction model based on the multiple sets of training data pairs.

[0011] Preferably, for each pressure distribution array, under the loaded state of the slotted counterweight, the laser displacement sensor passes through the slots on the slotted counterweight to perform in-situ measurements on the surface of the training crystal below each partition, obtaining the pre-processing loading surface shape data under the loaded state. This includes: controlling a two-dimensional moving platform to sequentially move the laser displacement sensor to a predetermined position above each slotted counterweight; driving the laser displacement sensor to scan along the center line of each scanning direction of the herringbone-shaped slot on the slotted counterweight; recording the planar coordinates of multiple measurement points and the corresponding crystal surface height values ​​in each scanning direction according to a preset sampling interval; and storing the height data of each measurement point in association with the corresponding planar coordinates to form the pre-processing loading surface shape data under the loaded state.

[0012] Preferably, the process involves acquiring the unloaded global surface shape data of the training crystal after processing and comparing it with the unloaded training surface shape data. Based on the comparison result, it is determined whether the training crystal has undergone permanent deformation. This includes: acquiring the unloaded global surface shape data of the training crystal after processing; subtracting the height values ​​at the same coordinate points in the unloaded global surface shape data and the unloaded global surface shape data to obtain multiple permanent deformation distributions; if the absolute value of the height difference at any coordinate point in any of the multiple permanent deformation distributions exceeds a preset height difference threshold, then it is determined that the training crystal has undergone permanent deformation.

[0013] Preferably, the preset height difference threshold is set according to one or more of the following methods: calculated and set according to a preset ratio based on the final surface shape accuracy requirements of the crystal to be processed; set according to a multiple of the standard deviation of the measurement repeatability accuracy of the laser displacement sensor and the interferometer; or set according to three times the standard deviation of the statistical distribution of the unloaded surface shape height difference in multiple loading and unloading experiments.

[0014] Preferably, when it is determined that the training crystal has not undergone permanent deformation, the elastic deformation and material removal amount are calculated based on the unloaded training surface data, the pre-processing loaded surface data, and the post-processing loaded surface data, and training data pairs are formed. This includes: when it is determined that the training crystal has not undergone permanent deformation, subtracting the height values ​​at the same coordinate points in the pre-processing loaded surface data and the unloaded training surface data to obtain the elastic deformation distribution under the pressure distribution, and calculating the average value of the elastic deformation in each partition as the deformation feature value of each partition; subtracting the height values ​​at the same coordinate points in the pre-processing loaded surface data and the post-processing loaded surface data to obtain the material removal amount corresponding to each scanning direction; calculating the average removal depth of each partition based on the removal amount data of the slots in each direction within each partition, as the removal feature value of the partition; associating the deformation feature value and the removal feature value of each partition to form training data pairs, wherein the training data pairs are associated with the pressure, elastic deformation, and material removal amount of the slotted counterweight.

[0015] The second objective of this invention is to provide a slotted counterweight block, which is applicable to the crystal counterweight weight configuration method described above. The slotted counterweight block is provided with a cross-shaped groove running through its top and bottom, and the cross-shaped groove includes four scanning directions: 0°, 45°, 90°, and 135°.

[0016] This invention achieves accurate identification of surface unevenness by acquiring the global surface shape of the crystal under no load and extracting local high-point features. Simultaneously, by using the global lowest point as the reference height, it ensures that the expected removal amount for all zones is non-negative, fundamentally avoiding over-polishing of low points. Furthermore, it introduces a pre-trained pressure-deformation-removal amount correlation prediction model. This model considers both pressure and elastic deformation information, accurately reflecting the nonlinear relationship between contact state and material removal, thus outputting high-precision zone pressure requirements and directly mapping them to the weight configuration scheme of the slotted counterweight. Compared to traditional overall pressure or single counterweight methods, this scheme achieves independent quantitative pressure application to high and low regions, significantly improving the uniformity of material removal and surface shape convergence efficiency, greatly reducing process trial-and-error costs, and is particularly suitable for controllable flexible polishing of large-size, high-precision crystals. Attached Figure Description

[0017] 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 the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the crystal counterweight configuration method provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of polishing provided in an embodiment of the present invention; Figure 3 This is the slotted counterweight block provided in the embodiments of the present invention. Attached image description: 1. Polishing disc; 2. Grooved counterweight; 21. Cross-shaped groove; 3. Crystal; 4. Laser displacement sensor; 5. Two-dimensional moving platform. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 This invention provides a method for configuring the weight of a crystal counterweight, comprising: S101. Obtain the unloaded global surface shape data of the crystal to be processed, and extract surface shape features from the unloaded global surface shape data; S102. Divide the surface of the crystal to be processed into several partitions according to the surface features; S103. Determine the height value of the global minimum point from the unloaded global surface data as the polishing reference height, and determine the expected material removal amount of each partition based on the local high point of each partition and the polishing reference height. S104. Input the expected material removal amount of each partition into the pre-trained correlation prediction model to obtain the required pressure value of each partition. S105. Based on the pressure value required for each partition, determine the slotted counterweight block corresponding to each partition and form a counterweight block array configuration scheme.

[0022] This embodiment uses a KDP crystal (400mm×400mm aperture) as an example, but this solution is also applicable to the polishing of precision optical components such as sapphire, silicon wafers, and silicon carbide.

[0023] In this embodiment, by acquiring the global surface shape of the crystal without load and extracting local high-point features, accurate identification of the surface unevenness distribution is achieved. At the same time, using the global lowest point as the reference height ensures that the expected removal amount of all zones is non-negative, fundamentally avoiding over-polishing of low points. Moreover, it introduces a pre-trained pressure-deformation-removal amount correlation prediction model, which considers both pressure and elastic deformation information, and can truly reflect the nonlinear relationship between contact state and material removal, thereby outputting high-precision zone pressure requirements and directly mapping them to the weight configuration scheme of the slotted counterweight. Compared with the traditional overall pressure or single counterweight method, this scheme achieves independent quantitative pressure application to high and low regions, significantly improving the uniformity of material removal and surface shape convergence efficiency, greatly reducing the process trial and error cost, and is especially suitable for the controllable flexible polishing of large-size, high-precision crystals.

[0024] In this embodiment, step S101 involves acquiring unloaded global surface shape data of the crystal to be processed and extracting surface shape features from the unloaded global surface shape data. This includes: measuring the entire surface of the crystal to be processed using an interferometer to obtain unloaded global surface shape data of the crystal to be processed and its height data matrix; generating a PV value cloud map based on the height data matrix; and extracting surface shape features from the PV value cloud map, wherein the surface shape features include the position, amplitude, and distribution range of local high points.

[0025] In this embodiment, the interferometer achieves sub-nanometer measurement accuracy, enabling precise identification of microscopic undulations on the crystal surface. The crystal to be processed is placed on the measuring stage of the interferometer (such as ZygoVerifire), and the crystal surface is adjusted to be perpendicular to the interferometer's optical path. The interferometer is then activated to scan the entire crystal surface, obtaining a three-dimensional height data matrix with a size of 1024×1024×1024 and a spatial resolution of 0.4 mm / pixel. This three-dimensional height data matrix represents the unloaded global surface shape data of the crystal to be processed.

[0026] In this embodiment, the maximum and minimum values ​​of the height matrix are calculated to obtain peak and valley values. The height value of each pixel is mapped to a color (red represents a high point, and blue represents a low point) to generate a PV value cloud map. This cloud map visually displays the uneven distribution of the crystal surface.

[0027] In this embodiment, a sliding window (10mm × 10mm) is used to search for local maxima on the PV value cloud map. For each candidate point, its height difference with other points within the surrounding window is compared. If the difference is greater than the peak-to-valley value / 10, it is marked as a local high point. The planar coordinates, amplitude, and distribution range of each local high point (e.g., a circular area with a radius of 15mm centered on the high point) are recorded to form surface features. The PV value cloud map and local high point extraction provide a quantitative basis for subsequent partitioning and pressure configuration.

[0028] Specifically, amplitude refers to the difference between the height of a local high point and the minimum height of the three-dimensional height data matrix.

[0029] Specifically, the sliding window is 10mm x 10mm in size.

[0030] In this embodiment, step S102, dividing the surface of the crystal to be processed into several partitions according to the surface features, includes: extracting all local high points of the crystal to be processed from the surface features; dividing the area where the local high points with amplitudes exceeding a preset amplitude threshold are located into independent partitions, and dividing the remaining areas according to a preset grid to obtain several partitions, wherein the total number of partitions does not exceed a preset upper limit.

[0031] In this embodiment, 20% of the PV value of the entire surface is taken from all the local high points obtained in the previous step as the preset amplitude threshold.

[0032] In this embodiment, all local high points are traversed. If the amplitude of a local high point is greater than a preset amplitude threshold, then that local high point and its surrounding circular area with a radius of 20mm are designated as an independent partition. The boundary shape of this partition is a square (side length 40mm), which is consistent with the bottom area of ​​the slotted counterweight block used later.

[0033] In this embodiment, areas not covered by independent partitions are uniformly divided using a fixed grid (e.g., a 6×6 grid, with each grid cell measuring 40mm×40mm). If a grid cell overlaps with an independent partition, the boundary of the independent partition is used as the reference.

[0034] Count the number of all partitions. If the number exceeds 36, merge adjacent high-point partitions with smaller amplitudes or increase the grid size (e.g., change to a 5×5 grid) to ensure that the total number of partitions is ≤36.

[0035] In this embodiment, step S103, which involves determining the height value of the global lowest point from the unloaded global surface data as the polishing reference height, and determining the expected material removal amount for each partition based on the local high point of each partition and the polishing reference height, includes: determining the height value of the global lowest point from the unloaded global surface data as the polishing reference height; extracting the highest height value of the local high point within each partition; and for each partition, using the difference between its highest height value and the polishing reference height as the expected material removal amount corresponding to the partition.

[0036] In this embodiment, the height data matrix is ​​scanned to find the minimum value, which is then used as the polishing reference height. This reference height ensures that all high points are above the target, while the lowest points are not processed.

[0037] In this embodiment, for each partition, the maximum value searched within the boundary of that partition is recorded as the highest height value of the local high point. If there is no local high point within a partition (e.g., a flat area), the average value of all points within that partition is taken as the highest height value of the local high point.

[0038] In this embodiment, if the high point of the partition is higher than the reference, the thickness to be removed is the difference between the two; if the high point of the partition is equal to the reference height, the removal amount is zero, indicating that the partition is not polished or is simply kept in place.

[0039] In this embodiment, in step S105, the expected material removal amount for each partition is input into the pre-trained correlation prediction model to obtain the required pressure value for each partition. Prior to this, the method includes: acquiring the unloaded training surface shape data of the training crystal and various pressure distribution arrays adapted to the training crystal; for each pressure distribution array, under the loaded state of the slotted counterweight, a laser displacement sensor passes through the slot on the slotted counterweight to perform in-situ measurements on the surface of the training crystal below each partition, obtaining the pre-processing loaded surface shape data under the loaded state; after the training crystal polishing process is completed, in-situ measurements are performed again under the condition that the loaded state of the slotted counterweight remains unchanged, obtaining the training crystal under the loaded state. After processing, load surface shape data is obtained; the global surface shape data of the training crystal after processing without load is acquired and compared with the training surface shape data without load. Based on the comparison results, it is determined whether the training crystal has undergone permanent deformation. When it is determined that the training crystal has not undergone permanent deformation, the elastic deformation and material removal amount are calculated based on the training surface shape data without load, the surface shape data before processing, and the surface shape data after processing with load, and training data pairs are formed. The training data pairs are associated with the pressure, elastic deformation, and material removal amount of the slotted counterweight. The weight combination of the slotted counterweight is changed multiple times to accumulate multiple sets of training data pairs, and a machine learning algorithm is used to train an associated prediction model based on the multiple sets of training data pairs.

[0040] In this embodiment, a KDP crystal of the same batch and size as the crystal to be processed is selected as the training crystal 3. Its surface is divided into partitions (e.g., 6×6) according to the same rules as the aforementioned steps. Multiple sets of slotted counterweights of different thicknesses are prepared: 4 mm, 6 mm, 8 mm, 10 mm, and 12 mm. Each thickness corresponds to a different pressure. At least five different pressure distribution arrays are designed (e.g., uniform pressure, high pressure in the center, high pressure at the edges, random distribution, etc.) to cover various working conditions.

[0041] In this embodiment, for each pressure distribution array, under the loaded state of the slotted counterweight, the laser displacement sensor passes through the slots on the slotted counterweight to perform in-situ measurements on the surface of the training crystal below each partition, obtaining the pre-processing loading surface shape data under the loaded state. This includes: controlling a two-dimensional moving platform to sequentially move the laser displacement sensor to a predetermined position above each slotted counterweight; driving the laser displacement sensor to scan along the center line of each scanning direction of the herringbone-shaped slot on the slotted counterweight; recording the planar coordinates of multiple measurement points and the corresponding crystal surface height values ​​in each scanning direction according to a preset sampling interval; and associating and storing the height data of each measurement point with the corresponding planar coordinates to form the pre-processing loading surface shape data under the loaded state.

[0042] Specifically, such as Figure 2 As shown, the laser displacement sensor 4 can be a Keyence LK-G500 with a resolution of 0.01μm. It is mounted on a two-dimensional moving platform 5, which is integrated into the polishing machine tool. The two-dimensional moving platform 5 has a travel of 500mm × 500mm and a repeatability of ±1μm. The platform is driven by a servo motor.

[0043] In this embodiment, for each pressure distribution array, the following operations are performed: First, the mobile platform is moved by the CNC system to be directly above the slotted counterweight in the first section (e.g., row 1, column 1), so that the center of the laser spot is aligned with the center of the star-shaped groove on the upper surface of the counterweight. The star-shaped groove includes straight grooves along four scanning directions: 0 degrees, 45 degrees, 90 degrees, and 135 degrees. The grooves are interconnected, with a width of 2 mm, and run through the entire thickness of the counterweight.

[0044] Next, the laser displacement sensor is driven to move forward along the center line of the groove in the 0-degree direction with a step size of 0.5 mm, recording the planar coordinates of each measurement point and the corresponding crystal surface height value. After reaching the end of the groove, it quickly retracts, and then switches sequentially to the 45-degree, 90-degree, and 135-degree directions, repeating the above scan. The information stored for each measurement point includes: the row and column coordinates of the partition, the scanning direction (0 degrees, 45 degrees, 90 degrees, or 135 degrees), the measurement point number in that direction, and the measured height value. The absolute planar coordinates of the measurement point are also recorded for precise alignment between subsequent different measurement data.

[0045] Then, the mobile platform moves to the next partition in sequence, repeating the above positioning and scanning operations until all partitions have been measured, and finally the pre-processing loading surface data is obtained.

[0046] In this embodiment, during the polishing process, crystal 3 is placed on the polishing disc 1 of the polishing machine, keeping the positions of all slotted counterweights 2 unchanged, and the polishing machine is started. The process parameters are set as follows: polishing disc rotation speed 60 revolutions per minute, polishing fluid is cerium dioxide suspension (10% concentration), and polishing time 30 minutes. During processing, moving crystal 3, slotted counterweights 2, or polishing disc 1 is prohibited to ensure the consistency of the measurement reference. After processing, the movement path and scanning trajectory in the pre-processing loaded surface shape data are reproduced, and the laser displacement sensor is driven again to obtain the post-loading height value of each partition in each direction, which is stored as post-processing loaded surface shape data.

[0047] In this embodiment, the process involves acquiring the unloaded global surface shape data of the training crystal after processing and comparing it with the unloaded training surface shape data. Based on the comparison result, it is determined whether the training crystal has undergone permanent deformation. This includes: acquiring the unloaded global surface shape data of the training crystal after processing; subtracting the height values ​​at the same coordinate points in the unloaded global surface shape data and the unloaded global surface shape data to obtain multiple permanent deformation distributions; if the absolute value of the height difference at any coordinate point in the multiple permanent deformation distributions exceeds a preset height difference threshold, it is determined that the training crystal has undergone permanent deformation.

[0048] In this embodiment, all slotted counterweights are removed from the training crystal, and the crystal is placed back into the interferometer to measure its unloaded global surface shape, which is recorded as the processed unloaded global surface shape data. This data is subtracted from the height values ​​of the same coordinate points in the initial unloaded global surface shape data to obtain multiple permanent deformation distributions. If the absolute value of the height difference at any coordinate point in any of the multiple permanent deformation distributions exceeds a preset height difference threshold (e.g., 5 nanometers), it is determined that the pressure distribution array has caused permanent deformation of the training crystal. If permanent deformation occurs, all data under this pressure distribution are discarded, and the maximum counterweight thickness in subsequent experiments is reduced (e.g., from 12 mm to 10 mm), and training for this pressure range is repeated. If no deformation occurs, the subsequent steps continue.

[0049] Specifically, the preset height difference threshold is set according to one or more of the following methods: calculated and set according to a preset ratio based on the final surface shape accuracy requirements of the crystal to be processed; set according to a multiple of the standard deviation of the measurement repeatability accuracy of the laser displacement sensor and interferometer; or set according to three times the standard deviation of the statistical distribution of the unloaded surface shape height difference in multiple loading and unloading experiments.

[0050] In this embodiment, when it is determined that the training crystal has not undergone permanent deformation, the elastic deformation and material removal are calculated based on the unloaded training surface data, the pre-processing loaded surface data, and the post-processing loaded surface data, and training data pairs are formed. This includes: when it is determined that the training crystal has not undergone permanent deformation, subtracting the height values ​​at the same coordinate points in the pre-processing loaded surface data and the unloaded training surface data to obtain the elastic deformation distribution under the pressure distribution, and calculating the average value of the elastic deformation in each partition as the deformation characteristic value of each partition; subtracting the height values ​​at the same coordinate points in the pre-processing loaded surface data and the post-processing loaded surface data to obtain the material removal amount corresponding to each scanning direction; calculating the average removal depth of each partition based on the material removal amount in the four scanning directions within each partition, as the removal characteristic value of the partition; associating the deformation characteristic value and the removal characteristic value of each partition to form training data pairs, wherein the training data pairs are associated with the pressure, elastic deformation, and material removal amount of the slotted counterweight.

[0051] In this embodiment, the pre-processing loaded surface shape, the post-processing loaded surface shape, and the initial unloaded global surface shape are aligned to the same coordinate system by coordinate matching (using the absolute planar coordinates of each measuring point as the primary key).

[0052] First, the elastic deformation is calculated. For each measurement point, the height of the loaded surface before processing is subtracted from the height of the initial unloaded global surface to obtain the elastic deformation at that point. For each partition, the arithmetic mean of the elastic deformation of all measurement points within that partition (i.e., all measurement points in the four scanning directions) is taken as the characteristic value of the deformation of that partition.

[0053] Then, the material removal amount is calculated. For each measurement point, the height of the loaded surface before processing is subtracted from the height of the loaded surface after processing to obtain the removal depth at that point. For each partition, the line-average removal depth for each of the four scanning directions is calculated separately, which is the sum of the removal depths of all measurement points on a measurement line divided by the number of measurement points. Next, the arithmetic mean of the line-average removal depths of the four scanning directions is calculated to obtain the overall average removal depth of the partition, which is used as the removal amount characteristic value of the partition.

[0054] Finally, the pressure value of each zone is calculated from the weight of the counterweight, and the deformation characteristic value and the material removal characteristic value are associated and stored to form a training data pair. The training data pair is associated with the pressure, elastic deformation and material removal of the slotted counterweight.

[0055] Specifically, the pressure value of each zone is calculated from the weight of the counterweight.

[0056] In this embodiment, the weight combination of the slotted counterweights is changed multiple times (i.e., the pressure distribution array is changed), and the aforementioned steps for forming training data pairs are repeated to accumulate multiple sets of training data pairs. In this embodiment, a total of 20 different pressure distribution experiments are conducted, with 36 partition samples obtained in each experiment, for a total of 720 samples.

[0057] In this embodiment, all samples are categorized according to their partition locations to form a training dataset. The input features for each sample are the pressure value and deformation value, and the output label is the removal amount value. The samples are randomly divided into a training set and a validation set in a 4:1 ratio.

[0058] A three-layer fully connected neural network was selected, with two nodes in the input layer representing pressure and deformation, and 64 nodes in each hidden layer using ReLU activation. The output layer had one node representing the amount of material removed. The mean squared error loss function was used, and the Adam algorithm was employed as the optimizer with a learning rate of 0.001. Backpropagation training was performed on the training set for 200 epochs, with 32 samples per batch. After each epoch, the mean squared error was calculated on the validation set. If the validation loss did not decrease for 10 consecutive epochs, training was stopped early, and the model with the minimum validation loss was saved. The final result is a neural network mapping where the amount of material removed is equal to the value of pressure and deformation as independent variables.

[0059] Alternatively, for small sample cases, a multinomial regression model can be used, such as a linear combination of the removal amount equal to the pressure, deformation amount, and their quadratic and cross terms, with the coefficients solved by the least squares method.

[0060] In this embodiment, during training or after predicting a new crystal, a global index can be further calculated. The sum of the removal feature values ​​of all partitions is divided by the total number of partitions to obtain the global average removal depth. The sum of the squares of the differences between the removal feature values ​​of all partitions and the global average removal depth is then divided by the total number of partitions, and the square root is taken to obtain the global removal uniformity index (i.e., standard deviation). Furthermore, for each partition, the standard deviation of the average removal depth in the four scanning directions can be calculated as the uniformity index within that partition, and the maximum value of this index among all partitions is identified to pinpoint the local region with the most severe anisotropy.

[0061] In this embodiment, the removal amount characteristic value of each partition can be marked with different colors on the crystal planar diagram to form a removal amount cloud map; the uniformity index within each partition can be marked with different colors to form a uniformity distribution map. Comparing the removal amount cloud map with the PV value cloud map can intuitively verify whether the high point area has been effectively removed; the uniformity distribution map highlights areas with severe anisotropy, which can be used to guide the next round of process optimization, such as adjusting the placement of counterweights or polishing parameters.

[0062] In this embodiment, for a new crystal to be processed, its initial unloaded global surface shape is first obtained and divided into partitions, and the expected material removal amount for each partition is calculated. Then, for each partition, the pressure value needs to be solved so that the model's predicted removal amount equals the expected removal amount. Since the correlation prediction model is a positive mapping, and the deformation characteristic value is related to the pressure value (an approximately linear relationship can be obtained by offline measurement of the crystal stiffness curve), a numerical optimization method (e.g., Bayesian optimization) is used to search for the pressure value within the pressure range that minimizes the absolute difference between the model's predicted removal amount and the expected removal amount. For polynomial models, the equations can be solved directly.

[0063] After obtaining the required pressure values ​​for each zone, the required mass of the slotted counterweight is calculated based on the relationship between pressure and the weight of the counterweight (pressure multiplied by the base area divided by gravitational acceleration). The counterweight closest to the calculated mass is selected from a pre-manufactured thickness sequence. Finally, a configuration table or two-dimensional layout diagram is output, indicating the thickness or weight of the counterweight to be placed in each zone, forming a counterweight array configuration scheme.

[0064] Understandably, the thickness of the slotted counterweight is set according to the actual situation. In this embodiment, the thickness sequence of the slotted counterweight is 4 mm, 6 mm, 8 mm, 10 mm, and 12 mm.

[0065] like Figure 3As shown, this embodiment of the invention also provides a slotted counterweight block, which is applicable to the crystal counterweight block weight configuration method described above. The slotted counterweight block is provided with a cross-shaped groove running through its top and bottom, and the cross-shaped groove includes four scanning directions: 0°, 45°, 90°, and 135°.

[0066] In this embodiment, the slotted counterweight is made of stainless steel (SUS304) with a density of 7.93 grams per cubic centimeter. It is a cube with a base length of 40 mm and various height (thickness) specifications including 4 mm, 6 mm, 8 mm, 10 mm, and 12 mm, with a thickness tolerance of ±0.05 mm. A cross-shaped groove is machined through the upper and lower surfaces of the counterweight. This groove consists of four straight grooves along the 0°, 45°, 90°, and 135° directions, intersecting at the center of the counterweight. The groove width is 2.0 mm, with a tolerance of ±0.05 mm. The ends of the grooves are rounded (0.5 mm radius) to reduce stress concentration. The bottom surface in contact with the crystal is precision ground, with a flatness better than 5 micrometers.

[0067] In use, according to the configuration, slotted counterweights of corresponding thickness are placed on designated sections of the crystal surface, applying pressure under their own weight. During measurement, the laser displacement sensor's beam passes through the star-shaped slots from directly above and illuminates the crystal surface. Because the slot width is larger than the spot diameter (approximately 0.5 mm), the beam is unobstructed. The slots in four scanning directions allow the sensor to scan at different angles, thereby acquiring morphological data in multiple directions within the section for evaluating the isotropic removal effect.

[0068] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for configuring the weight of a crystal counterweight, characterized in that, include: Obtain the unloaded global surface shape data of the crystal to be processed, and extract surface shape features from the unloaded global surface shape data; The surface of the crystal to be processed is divided into several zones based on the surface features described above; The height value of the global minimum point is determined from the unloaded global surface data and used as the polishing reference height. Based on the local high point of each partition and the polishing reference height, the expected material removal amount of each partition is determined. The expected amount of material to be removed from each partition is input into a pre-trained correlation prediction model to obtain the required pressure value for each partition. Based on the required pressure values ​​for each partition, determine the slotted counterweights for each partition and form a counterweight array configuration scheme.

2. The method for configuring the weight of the crystal counterweight according to claim 1, characterized in that, The process of acquiring unloaded global surface shape data of the crystal to be processed and extracting surface shape features from the unloaded global surface shape data includes: The entire surface of the crystal to be processed is measured using an interferometer to obtain the unloaded global surface shape data and its height data matrix of the crystal to be processed. Generate a PV value cloud map based on the height data matrix; The surface features are extracted from the PV value cloud map, and the surface features include the location, magnitude and distribution range of local high points.

3. The method for configuring the weight of the crystal counterweight according to claim 2, characterized in that, The process of dividing the surface of the crystal to be processed into several partitions based on the surface features includes: Extract all local high points of the crystal to be processed from the surface features; The region containing local high points whose amplitude exceeds a preset amplitude threshold is divided into independent partitions, and the remaining regions are divided according to a preset grid to obtain several partitions. The total number of partitions does not exceed a preset upper limit.

4. The method for configuring the weight of the crystal counterweight according to claim 1, characterized in that, The step of determining the height value of the global minimum point from the unloaded global surface data as the polishing reference height, and determining the expected material removal amount for each partition based on the local high point of each partition and the polishing reference height, includes: The height value of the global minimum point is determined from the unloaded global surface data and used as the polishing reference height; Extract the highest height value of local high points within each partition; For each partition, the difference between its highest height value and the polishing reference height is taken as the expected amount of material to be removed for that partition.

5. The method for configuring the weight of the crystal counterweight according to claim 1, characterized in that, The expected material removal amount for each partition is input into a pre-trained association prediction model to obtain the required stress value for each partition. This process also includes: Acquire unloaded training surface shape data of the training crystal and various pressure distribution arrays adapted to the training crystal; For each pressure distribution array, under the slotted counterweight loading state, in-situ measurements are performed on the surface of the training crystal below each partition to obtain the pre-processing loading surface shape data under the loading state. After the training crystal polishing process is completed, in-situ measurements are performed again under the condition that the loading state of the slotted counterweight remains unchanged, to obtain the post-processing loading surface shape data of the training crystal under the loading state. Obtain the unloaded training surface shape data of the crystal after processing and compare it with the unloaded training surface shape data. Based on the comparison results, determine whether the crystal has undergone permanent deformation. When it is determined that the training crystal has not undergone permanent deformation, the elastic deformation and material removal are calculated based on the unloaded training surface data, the pre-processing loaded surface data, and the post-processing loaded surface data, and training data pairs are formed. The training data pairs are associated with the pressure, elastic deformation, and material removal of the slotted counterweight. The weight combination of the slotted counterweights was changed multiple times to accumulate multiple sets of training data pairs. A machine learning algorithm was then used to train an association prediction model based on the multiple sets of training data pairs.

6. The method for configuring the weight of the crystal counterweight according to claim 5, characterized in that, For each pressure distribution array, under the loaded state of the slotted counterweight, a laser displacement sensor passes through the slots on the slotted counterweight to perform in-situ measurements on the surface of the training crystal below each partition, obtaining the pre-processing loading surface shape data under the loaded state, including: The two-dimensional moving platform is controlled to move the laser displacement sensor sequentially to a predetermined position above each slotted counterweight; The laser displacement sensor is driven to scan along the center line of each scanning direction of the cross-shaped groove on the slotted counterweight; In each scanning direction, the planar coordinates of multiple measurement points and the corresponding crystal surface height values ​​are recorded according to a preset sampling interval; The height data of each measurement point is associated with and stored with the corresponding planar coordinates to form the pre-processing loading surface shape data under loading conditions.

7. The method for configuring the weight of the crystal counterweight according to claim 5, characterized in that, Obtain the unloaded global surface shape data of the training crystal after processing and compare it with the unloaded training surface shape data. Based on the comparison results, determine whether the training crystal has undergone permanent deformation, including: Obtain the unloaded global surface shape data of the crystal used for training; By subtracting the height values ​​at the same coordinate points from the unloaded global surface shape data of the crystal after processing, and the unloaded global surface shape data, multiple permanent deformation distributions are obtained. If, in any of the multiple distributions of permanent deformation, the absolute value of the height difference at any coordinate point exceeds a preset height difference threshold, then the training crystal is determined to have undergone permanent deformation.

8. The method for configuring the weight of the crystal counterweight according to claim 7, characterized in that, The preset height difference threshold is set according to one or more of the following methods: The final surface accuracy requirements of the crystal to be processed are calculated and set according to a preset ratio; The measurement repeatability of the laser displacement sensor and interferometer is set according to the multiple of the standard deviation. The standard deviation of the statistical distribution of the height difference of the unloaded surface was set to three times the standard deviation of the distribution of the height difference of the unloaded surface in multiple loading and unloading experiments.

9. The method for configuring the weight of the crystal counterweight according to claim 5, characterized in that, When it is determined that the training crystal has not undergone permanent deformation, the elastic deformation and material removal are calculated based on the no-load training surface data, the pre-processing loaded surface data, and the post-processing loaded surface data, and training data pairs are formed, including: When it is determined that the training crystal has not undergone permanent deformation, the height values ​​at the same coordinate points in the pre-processed loaded surface data and the unloaded training surface data are subtracted to obtain the elastic deformation distribution under the pressure distribution, and the average value of the elastic deformation in each partition is calculated as the deformation characteristic value of each partition. Subtract the height values ​​at the same coordinate points from the pre-processing loading surface data and the post-processing loading surface data to obtain the material removal amount corresponding to each scanning direction; Based on the amount of material removed in the four scanning directions within each partition, the average removal depth of the partition is calculated and used as the removal amount characteristic value of the partition. The deformation feature value and the material removal feature value of each partition are associated to form a training data pair, which is associated with the pressure, elastic deformation and material removal of the slotted counterweight.

10. A slotted counterweight, characterized in that, The slotted counterweight is applicable to the crystal counterweight weight configuration method as described in any one of claims 1-9. The slotted counterweight is provided with a cross-shaped groove running through its top and bottom, and the cross-shaped groove includes four scanning directions: 0°, 45°, 90°, and 135°.

Images