Intelligent optimization method for ultra-precision machining process based on cutting vibration coupling simulation

Abstract

The invention discloses an intelligent optimization method for an ultra-precision machining process based on cutting vibration coupling simulation. The intelligent optimization method comprises the following steps: S1, establishing a workpiece global multi-physics field coupling simulation model; s2, dynamic cutting force of the whole workpiece cutting process is obtained; s3, the cutting force serves as excitation to be input into the kinetic model of the machine tool-tool-workpiece system, and the dynamic displacement response of the tool relative to the workpiece is obtained; s4, superposing the response into the cutting simulation model; s5, performing stress release process simulation; s6, surface topography accurate reconstruction based on improved multi-stage screening; s7, calculating surface shape error distribution based on the reconstructed surface morphology; s8, constructing a deep neural network prediction model; and S9, setting an optimization target value and a constraint condition, and searching an optimal solution in a process parameter space. According to the method, the optimal process parameter combination meeting the performance requirement can be quickly found, so that the trial cutting frequency is reduced, and the process debugging period is shortened.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of ultra-precision machining, and particularly relates to an intelligent optimization method for ultra-precision machining process parameters of large plane optical elements based on cutting force excitation coupling simulation and deep learning, which is suitable for pre-evaluation of machining quality and optimization guidance of process parameters of high-precision optical elements in the fields of aerospace and the like. BACKGROUND

[0002] Large plane optical elements (such as space telescope main mirrors and the like) are core devices of modern high-end optical systems, and the surface shape precision thereof directly affects the performance of the entire optical system. Such elements usually have the characteristics of large size (the diameter can reach several meters), high precision (the surface shape precision is required to be more than λ / 50, wherein λ=632.8 nm), and low roughness (Ra<0.5 nm), and thus put forward extremely high requirements on ultra-precision machining technology.

[0003] However, due to the large size of the elements, the machine tool geometric errors, thermal deformation, vibration in the machining process, and the complex mechanical behavior of the workpiece-clamp system will generate surface shape errors on the element surface, which seriously affect the wavefront quality and imaging performance of the optical element. For example, in the clamping-cutting process, a certain stress accumulation will be generated on the workpiece, and after the stress is released, a deformation will be generated to deviate from the cutting target. At the same time, the dynamic cutting force generated in the cutting will excite the vibration of the machine tool-tool-workpiece system, and such vibration will leave marks on the element surface, affecting the optical performance.

[0004] Traditional ultra-precision machining process parameter selection mainly relies on experience and trial cutting, and often needs an iterative cycle of “machining-measuring-compensation”, which is low in efficiency and high in cost, and it is difficult to predict the final surface shape before machining. The traditional method is difficult to establish a quantitative relationship from the process parameters (cutting speed, feed rate, cutting depth, etc.) to the final optical performance (surface shape precision, surface roughness, etc.), and the process selection lacks scientific basis. The existing methods mostly use empirical parameters or simple optimization algorithms, and lack intelligent process parameter optimization technology based on big data and artificial intelligence.

[0005] Finite element simulation technology is an important means for analyzing the ultra-precision machining process, and can predict the stress distribution, material deformation and surface quality in the machining process. However, the existing finite element simulation mostly adopts static or quasi-static analysis, ignores the feedback influence of machine tool vibration excited by cutting force on the cutting process, and cannot accurately predict the dynamic cutting behavior in actual machining. At the same time, after the cutting simulation is completed, the workpiece surface is mixed with a large number of cutting chip grids and failure grids, and the traditional manual screening method is extremely low in efficiency, and the post-processing technology is low in accuracy and efficiency in surface topography reconstruction.

[0006] The patent application with the publication number CN119238283A discloses a deterministic cutting compensation processing method for infrared thin-wall optical elements, which is based on deep learning for residual stress prediction and deterministic compensation, and mainly focuses on the processing compensation strategy affected by residual stress. The patent application with the publication number CN118690615A discloses a part milling process simulation method considering multi-step stress release process, which focuses on the simulation modeling process of multi-step simulation. The patent application with the publication number CN111144040A discloses a global deformation simulation method suitable for plane milling multi-step process, which focuses on the simulation analysis algorithm method. However, the above methods do not solve the simulation post-processing data flow problem, and do not involve cutting vibration coupling analysis and intelligent process optimization.

[0007] In summary, therefore, there is an urgent need for an intelligent process optimization method based on cutting vibration coupling simulation and deep learning, which can meet the processing method of ultra-precision optical elements. SUMMARY

[0008] The present application aims to solve the technical problems of lack of cutting vibration coupling analysis, unclear relationship between process parameters and optical performance, and low intelligence in the existing ultra-precision machining process, and provides an ultra-precision machining process optimization method for large plane optical elements based on cutting vibration coupling simulation and deep learning.

[0009] The purpose of the present application is achieved by the following technical scheme: an ultra-precision machining process intelligent optimization method based on cutting vibration coupling simulation, comprising the following steps:

[0010] S1, establishing a global multi-physical field coupling simulation model of the workpiece: establishing a three-dimensional finite element model of the large plane optical element containing the whole process of "clamping-cutting-stress release", and constructing a dynamic model of the machine tool-tool-workpiece system;

[0011] S2, based on the "clamping-cutting-stress release" model, setting the process parameters, and obtaining the dynamic cutting force Fc(t) of the whole cutting process of the workpiece;

[0012] S3, inputting the cutting force Fc(t) as the excitation into the dynamic model of the machine tool-tool-workpiece system for simulation, and obtaining the dynamic displacement response δ(x,y,t) of the tool relative to the workpiece;

[0013] S4, superimposing the dynamic displacement response δ(x,y,t) obtained in step S3 into the three-dimensional finite element model of the large plane optical element containing the whole process of "clamping-cutting-stress release", and obtaining the cutting process simulation result considering the vibration coupling effect;

[0014] S5, after completing the cutting simulation, remove the cutting and clamping load, and perform a stress release process simulation to obtain the workpiece stress field and displacement field distribution in the final stable state;

[0015] S6, surface topography accurate reconstruction based on improved multi-stage screening: read data and perform preprocessing, filter invalid grids, then intelligently identify and remove chip grids, then preliminarily screen surface possible nodes according to spatial distribution, and again screen through local extreme positioning and fine screening; finally, calculate the coordinates after deformation, and reconstruct the continuous surface;

[0016] S7, calculate the surface error distribution based on the reconstructed surface topography, and compare it with the ideal plane to extract high-frequency, medium-frequency and low-frequency error components; calculate the surface accuracy, surface roughness and wavefront error as key optical performance indicators;

[0017] S8, generate sample combinations in the process parameter space, and perform simulation calculation through steps S1-S7 to form a "process parameter-optical performance" sample pair dataset; preprocess the data, extract process parameter features and optical performance features; construct a deep neural network prediction model, and train the deep neural network prediction model using the supervised learning paradigm;

[0018] S9, according to the application requirements of the optical element, set the surface accuracy and surface roughness as the optimization target value, and set the equipment capacity, processing efficiency and tool life as the constraint condition, search for the optimal solution in the process parameter space, and output the optimal process parameter combination and confidence evaluation that meet the optical performance requirements.

[0019] In the step S6, the surface topography reconstruction includes the following steps:

[0020] S6.1, data preprocessing and initialization: read the finite element simulation data file, locate the analysis step and time frame corresponding to the completion of cutting; extract the spatial coordinates, displacement field and state information of all grid elements and nodes; establish the node-element association topology relationship and construct the data index table;

[0021] S6.2, filter invalid grids: first, remove invalid elements and associated nodes according to element state values, and count the remaining valid nodes and the number of valid element connections for each node;

[0022] S6.3, intelligent identification and removal of chip grids: calculate the displacement component of each node in the cutting direction, set a dynamic displacement threshold Δd; mark and remove the nodes with displacement components exceeding the threshold as chips;

[0023] S6.4, Space distribution quantization screening suspected surface nodes: Establish a spatial coordinate sequence along the cutting direction, extract the top a% nodes as candidate surface nodes; Use a spatial clustering algorithm to identify and separate independent surface regions;

[0024] S6.5, Local extremum positioning and fine screening: Divide the remaining nodes into multiple sub-grids according to the minimum grid size multiplied by β; In each sub-grid, search for local extremum along the cutting direction, and extract the local optimal node as the final surface node;

[0025] S6.6, Surface reconstruction and optimization: Extract the original coordinates and deformation displacement of the final surface nodes; Calculate the node deformation coordinate X_new: X_new = X_origin + U_displacement, X_origin is the coordinate before deformation, and U_displacement is the deformation displacement; Use surface fitting to generate a continuous surface topography.

[0026] The beneficial effects of the present application are: The present application establishes a quantitative mapping relationship between process parameters and optical performance, and through intelligent optimization algorithm, the optimal process parameter combination meeting the performance requirements can be quickly found, thereby reducing the number of trial cuts and shortening the process debugging period. At the same time, in the simulation of ultra-precision machining, the cutting force excitation response analysis is introduced, a vibration-cutting coupling model is established, and the influence mechanism of vibration on machining quality is revealed, thereby providing a theoretical basis for vibration suppression and process parameter optimization.

[0027] The automatic multi-level screening method of cutting simulation post-processing can accurately locate the surface nodes after cutting simulation, and only needs a few minutes to process simulation models with hundreds of thousands of nodes, thereby significantly improving the finite element post-processing time. It can provide scientific process selection basis for high-end optical manufacturing, improve the stability and consistency of machining quality, and reduce research and development cost and production risk. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The overall technical flowchart of the present application is shown in the figure;

[0029] Figure 2 The clamping-cutting-stress release finite element simulation flowchart is shown in the figure;

[0030] Figure 3 The multi-level screening surface reconstruction algorithm flowchart is shown in the figure;

[0031] Figure 4 The process optimization deep learning schematic diagram is shown in the figure. DETAILED DESCRIPTION

[0032] The technical solutions of the present application are further described below in conjunction with the drawings.

[0033] In this embodiment, the polycrystalline copper plane mirror ultra-precision fly-cutting machining with the size of 300*130*10mm is taken as an example to illustrate the implementation process of the present application. The surface accuracy is required to be PV≯0.25λ, RMS≯0.05λ, and the surface roughness is required to be less than 0.5nm. The machining equipment is an ultra-precision fly-cutting machine tool, the maximum spindle speed is 300rpm, and the feeding accuracy is 0.01μm.

[0034] As shown in Figure 1 The cutting vibration coupling simulation-based ultra-precision machining process intelligent optimization method of the present application comprises the following steps:

[0035] S1, a global multi-physical field coupling simulation model of a workpiece is established;

[0036] S1.1, a three-dimensional model containing the workpiece, the fixture and the tool is established, a three-dimensional finite element model of a large plane optical element containing the whole process of "clamping-cutting-stress release" is established, the simulation parameters such as the material constitutive relation of polycrystalline copper, the boundary conditions and the contact conditions are established according to the Johnson-cook model, and the clamping-cutting-stress release finite element simulation process of the present embodiment is as shown in Figure 2 In the present embodiment, the workpiece grid node is about 76,753; the material constitutive of polycrystalline copper and the damage model are established according to the Johnson-cook model, and the damage displacement is set to be the minimum grid length;

[0037] S1.2, a dynamic model of the machine tool-tool-workpiece system is constructed, including the modal parameters of the spindle system, the feeding system and the support system; the first order natural frequency of the spindle system is 480Hz, and the damping ratio is 0.05;

[0038] S2, based on the "clamping-cutting-stress release" model, the process parameters are set, in the present embodiment, the spindle speed n=280rpm, the feeding amount f=2μm / rev, the cutting depth ap=5μm, and the cutting speed v=π*tool diameter D*spindle speed n / 60; the dynamic cutting force Fc(t) of the whole cutting process of the workpiece is obtained, in the present embodiment, the maximum value of the dynamic cutting force Fc(t) is 2.5N;

[0039] S3, the cutting force Fc(t) is taken as an excitation and input into the dynamic model of the machine tool-tool-workpiece system for simulation, the vibration response analysis of the machine tool-tool-workpiece system is carried out, and the dynamic displacement response δ(x,y,t) of the tool relative to the workpiece is obtained; in the present embodiment, the spindle radial vibration amplitude is 0.008μm (20Hz) and 0.015μm (480Hz resonance), and the workpiece axial vibration amplitude is 0.012μm (85Hz) and 0.028μm (320Hz); the main vibration frequency components are identified to be 20Hz, 85Hz, 320Hz and 480Hz.

[0040] S4, considering the influence of vibration on the actual cutting depth and cutting force, superimposing the dynamic displacement response δ(x, y, t) obtained in step S3 to the large plane optical element three-dimensional finite element model containing the "clamping-cutting-stress release" whole process, correcting the tool path in the cutting model simulation, combining the adaptive mesh technology to solve the multi-scale of vibration displacement and macro feed, carrying out the cutting simulation considering the influence of tool vibration, and obtaining the cutting process simulation result considering the coupling effect of vibration.

[0041] S5, after completing the cutting simulation, removing the cutting and clamping load, and simulating the stress release process, considering the influence of the step-by-step release process of the clamping constraint on the workpiece deformation, and obtaining the stress field and displacement field distribution of the workpiece in the final stable state; in this embodiment, the maximum convex deformation of the center area is 18.5nm, and the edge area sinks 12.3nm; the residual stress distribution: the maximum residual stress is 32.8MPa, mainly concentrated in the clamping contact area.

[0042] S6, accurate surface topography reconstruction based on improved multi-level screening: the workpiece blank warping model, first through vacuum clamping and flattening, then surface cutting, and model springback after cutting. At this time, if the model surface nodes are manually selected, the operation is difficult. Script selection is difficult to define a single description of the node, so the method is applied to multi-level automatic screening of surface nodes. First, read the data and perform preprocessing and initialization, filter invalid grids, then intelligently identify and remove the chip grid, then preliminarily screen the surface possible nodes according to the spatial distribution, and again through local extreme positioning and fine screening; finally, calculate the deformed coordinates, and reconstruct the continuous surface by using cubic spline interpolation or NURBS fitting technology.

[0043] The workpiece blank warping model is first flattened by vacuum clamping, then surface cutting is performed, and the model rebounds after cutting. At this time, if the model surface nodes are manually selected, the operation is difficult. Script selection is difficult to define a single description of the node, so the method of the application is applied to multi-level automatic screening of surface nodes, and then surface reconstruction is performed. As shown in Figure 3 The surface topography reconstruction includes the following steps:

[0044] S6.1, data preprocessing and initialization: read the finite element simulation data file, locate the corresponding analysis step and time frame of cutting completion; extract the spatial coordinates, displacement field and state information of all grid elements and nodes; the specific method is: open the ABAQUS simulation output file (.odb format), locate to the Step-2, Frame-100 corresponding to the cutting completion time. The simulation model contains N=76,753 nodes. Extract the node coordinate matrix Node_Coord[Nx3] and the displacement field matrix Displacement[Nx3], establish the node-element association topology relationship, and construct the data index table.

[0045] S6.2, filtering of failed grids: according to the element state value, remove the failed elements and their associated nodes, and count the remaining valid nodes and the number of valid element connections for each node; read the element state value Status, remove the failed elements and their associated nodes with Status=0; in this embodiment, the failed elements and their associated 24,916 nodes, the remaining valid nodes are 51,837.

[0046] S6.3, intelligent identification and removal of chip grids: calculate the displacement component dz of each node in the cutting direction (Z direction), set the dynamic displacement threshold value Ad=k-d_avg(k is the adjustment coefficient, d_avg is the average displacement); mark and remove the nodes with displacement component exceeding the threshold value; the chip identification method further includes:

[0047] S6.3.1, adaptive threshold adjustment: dynamically adjust the displacement threshold value according to the material properties and cutting parameters;

[0048] S6.3.2, multi-dimensional discrimination: comprehensive judgment is made by combining the displacement components in X, Y and Z directions; in this embodiment, the threshold values in x and z directions are set as: Δx=30mm, Δz=10mm.

[0049] S6.3.3, when a single limit displacement threshold value cannot completely remove the chip, introduce the connection element number discrimination criterion: when the node connection element number N_conn<5 or the displacement exceeds the threshold value, and the proportion of such nodes exceeds 50%, it is determined as the chip area.

[0050] S6.3.4, neighborhood consistency test: neighborhood verification is performed on the nodes marked as chips to ensure the accuracy of the discrimination. In this embodiment, 402 nodes with displacement exceeding the threshold value are identified and preliminarily marked as chips, and these points are removed.

[0051] S6.4, Space distribution quantization screening suspected surface nodes: Establish spatial coordinate sequence along the cutting direction, extract the first a% (a is an adjustable parameter, 15% in this embodiment, 7,717 nodes are obtained) nodes as candidate surface nodes; Use spatial clustering algorithm to identify and separate independent surface regions; Space distribution screening further includes:

[0052] S6.4.1, Direction self-adaptation: Automatically determine the optimal screening direction of the machined workpiece according to the tool path;

[0053] S6.4.2, Density weight: Introduce node density weight to improve the screening accuracy of sparse areas;

[0054] S6.4.3, Boundary protection: Special protection of workpiece boundary nodes to avoid loss of boundary information.

[0055] S6.5, Local extremum positioning and fine screening: Divide the remaining nodes extracted in S6.4 into multiple sub-grids according to the minimum grid size multiplied by β (grid refers to the connection between nodes, and β is the refinement coefficient); In each sub-grid, search for extreme value along the cutting direction, and extract the local optimal node as the final surface node; Extremum positioning further includes:

[0056] S6.5.1, Multi-scale analysis: Perform multiple extremum searches with different grid sizes to improve robustness;

[0057] S6.5.2, Gradient constraint: Introduce surface gradient continuity constraint to ensure the physical reasonableness of the reconstructed surface.

[0058] In this embodiment, the minimum grid size h_min = 0.05mm, the grid is divided by 0.5xh_min = 0.025mm, a total of 13,046 sub-grids are generated. In each sub-grid, search for Z direction extremum, and extract 6,524 local optimal nodes as final surface nodes.

[0059] S6.6, Surface reconstruction and optimization: Extract the original coordinates and deformation displacement of the final surface nodes; Calculate the node deformed coordinate X_new: X_new = X_origin + U_displacement, X_origin is the coordinate before deformation, and U_displacement is the deformation displacement; Use cubic spline interpolation or NURBS technology to reconstruct continuous surface, and generate continuous surface topography.

[0060] S7, calculate the surface error distribution based on the reconstructed surface topography, and compare it with the ideal plane to extract high, medium and low frequency error components; calculate the surface accuracy (PV value, RMS value), surface roughness, wavefront error as key optical performance indicators; then perform surface quality evaluation, calculate the characteristic values such as roughness parameters Ra, Rz, etc. In this embodiment, the RMS value obtained by reconstruction differs from the actual measured value by <2nm, verifying the accuracy of the method.

[0061] The surface error analysis result of this embodiment is: PV value: 52.3nm (about λ / 12=0.083λ, meeting the requirement of PV≯0.25λ); RMS value: 14.7nm (about λ / 43=0.023λ, meeting the requirement of RMS≯0.05λ).

[0062] S8, generate sample combinations in the 25x4x3=300 process parameter space through Latin hypercube sampling, and perform simulation calculation through steps S1-S7 to form a "process parameter-optical performance" sample pair data set; preprocess the data, extract process parameter features (cutting parameters, tool geometry, material properties) and optical performance features (PV, RMS, Ra, etc.) by feature engineering; construct a deep neural network prediction model, including an input layer, a hidden layer and an output layer, the structure of which is as shown in Figure 4 The input layer inputs a multi-dimensional process parameter vector. The hidden layer adopts a multi-layer fully connected neural network architecture, with 3-5 layers and 64-256 neurons per layer; residual connection is introduced to solve the problem of deep network degradation; attention mechanism is added to automatically learn the weights of key process parameters; the output layer is the optical performance prediction value (PV, RMS, Ra, etc.). Supervised learning paradigm is adopted, mean square error loss function and Adam optimizer are used to train the deep neural network prediction model, cross-validation and regularization techniques are applied to prevent overfitting, and good prediction accuracy and generalization ability of the model are ensured.

[0063] In the training, the sample set is divided into training set, validation set and test set, each set accounts for 7:2:1, the batch size is 32, the learning rate is 0.001, and the iteration number is 1000; the total training time is about 4 hours (GPU acceleration), and the model training performance is: training set R²: 0.956; validation set R²: 0.921; test set R²: 0.918; average prediction error: 7.8%

[0064] S9、According to the application requirements of the optical element, set the surface shape accuracy and surface roughness as the optimization target values, and set the equipment capacity, processing efficiency, and tool life as the constraint conditions; in the optimization process, first, input all the process parameters of step S8 into the trained deep neural network prediction model to perform fast performance prediction, and remove the process parameters whose predicted performance does not meet the requirements, so as to avoid repeated simulation. Then, a genetic algorithm, a particle swarm algorithm, or a Bayesian optimization is used to search for an optimal solution in the process parameter space, and an optimal process parameter combination and a confidence evaluation that meet the optical performance requirements are output. In this embodiment, after about 2.5 hours of optimization calculation, a Pareto optimal solution set containing 23 process parameter combinations is obtained. According to the comprehensive score (considering the optical performance, processing efficiency, and tool life), an optimal parameter combination is selected. Actual processing verification is performed by using the optimized process parameters, and the obtained indexes are shown in Table 1.

[0065] Table 1

[0066] Performance indicator Deep learning predicted value Actual measured value Prediction error PV value (nm) 14.8 15.7 5.7% RMS value (nm) 4.1 4.4 6.8% Surface roughness Ra (nm) 0.39 0.41 4.8% Strehl ratio 0.85 0.82 3.5%

[0067] The actual processing result meets all the design requirements, and the effectiveness of the method of the present application is verified.

[0068] Those skilled in the art will appreciate that the embodiments described herein are presented for the purpose of helping the reader to understand the principles of the present application, and should be understood as not limiting the protection scope of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspirations disclosed in the present application, and these modifications and combinations do not deviate from the essence of the present application, and are still within the protection scope of the present application.

Claims

1. An intelligent optimization method for ultra-precision machining process coupled with cutting vibration simulation, characterized in that, The method comprises the following steps: S1, establishing a global multi-physical field coupling simulation model of the workpiece: establishing a large plane optical element three-dimensional finite element model including the whole process of "clamping-cutting-stress release", and constructing a dynamic model of the machine tool-tool-workpiece system; S2, based on the "clamping-cutting-stress release" model, setting process parameters, and obtaining the dynamic cutting force Fc(t) of the whole cutting process of the workpiece; S3, taking the cutting force Fc(t) as the excitation, inputting the dynamic model of the machine tool-tool-workpiece system for simulation, and obtaining the dynamic displacement response δ(x, y, t) of the tool relative to the workpiece; S4, superimposing the dynamic displacement response δ(x, y, t) obtained in step S3 to the large plane optical element three-dimensional finite element model including the whole process of "clamping-cutting-stress release", and obtaining the simulation result of the cutting process considering the vibration coupling effect; S5, after completing the cutting simulation, removing the cutting and clamping load, and performing stress release process simulation to obtain the stress field and displacement field distribution of the workpiece in the final stable state; S6, accurate reconstruction of surface topography based on improved multi-level screening: reading data and preprocessing, filtering invalid grids, then intelligently identifying and removing chip grids, then preliminarily screening possible surface nodes according to spatial distribution, and again screening through local extreme positioning and fine screening; finally, calculating the coordinates after deformation, and reconstructing the continuous surface; S7, calculating the surface shape error distribution based on the reconstructed surface topography, and comparing with the ideal plane to extract high-frequency, medium-frequency and low-frequency error components; calculating the surface roughness, wavefront error and surface roughness as key optical performance indicators; S8, generating sample combinations in the process parameter space, and performing simulation calculation through steps S1-S7 to form a "process parameter-optical performance" sample pair data set; preprocessing the data, extracting process parameter features and optical performance features; constructing a deep neural network prediction model, and training the deep neural network prediction model by using a supervised learning paradigm; S9, according to the application requirements of the optical element, setting the surface shape accuracy and surface roughness as the optimization target value, taking the equipment capacity, processing efficiency and tool life as the constraint condition, searching for the optimal solution in the process parameter space, and outputting the optimal process parameter combination and confidence evaluation which meet the optical performance requirements.

2. The super-precision machining process intelligent optimization method of cutting vibration coupling simulation according to claim 1, characterized in that, In step S6, the surface topography reconstruction comprises the following steps: S6.1, data preprocessing and initialization: reading the finite element simulation data file, locating the corresponding analysis step and time frame after cutting is completed; extracting the spatial coordinates, displacement field and state information of all grid elements and nodes; establishing the node-element association topology relationship, and constructing the data index table; S6.2, filtering invalid grids: first, removing invalid elements and associated nodes according to the element state value, and counting the remaining valid nodes and the number of valid element connections of each node; S6.3, intelligently identifying and removing chip grids: calculating the displacement component of each node in the cutting direction, setting a dynamic displacement threshold Δd; marking and removing the nodes with displacement component exceeding the threshold as chips; S6.4, Quantitative screening of suspected surface nodes: Establish spatial coordinate sequence along the cutting direction, extract the top α% nodes as candidate surface nodes; use spatial clustering algorithm to identify and separate independent surface regions; S6.5, Local extremum positioning and fine screening: Divide the remaining nodes into multiple sub-grids according to β times the minimum grid size; perform extremum search along the cutting direction in each sub-grid, and extract the local optimal nodes as the final surface nodes; S6.6, Surface reconstruction and optimization: Extract the original coordinates and deformation displacement of the final surface nodes; calculate the node coordinates after deformation X_new: X_new = X_origin + U_displacement, X_origin is the coordinate before deformation, U_displacement is the deformation displacement; use surface fitting to generate continuous surface topography.

Images