Geometric parameter collaborative optimization method for taper hole machining tool

Abstract

The invention relates to the technical field of collaborative optimization, in particular to a geometric parameter collaborative optimization method of a taper hole machining cutter, which comprises the following steps: by constructing a high-fidelity digital twin model, integrating multi-physics field coupling and machine tool dynamic characteristics based on a finite element method, generating a geometric parameter-performance data mapping set and training and calculating an agent model; outputting a Pareto solution set through multi-target global optimization; and constructing a constraint range based on the solution set, and calling a digital twin model to carry out local optimization to obtain an optimal geometric parameter combination. The method comprises a self-correction mechanism: correcting a material constitutive relation and a friction coefficient through experimental data; staged adaptive learning, NSGA-II and DBSCAN clustering are adopted, and the efficiency is optimized along with the method; and a Pareto stability index and transfer learning are introduced, so that the result robustness and the cross-task reusability are improved. According to the method, high-precision and high-efficiency geometric parameter collaborative optimization of the taper hole machining tool can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of collaborative optimization, in particular to a geometric parameter collaborative optimization method of a taper hole machining tool. BACKGROUND

[0002] In the prior art, for high-precision and high-finish taper hole machining, numerical control programming (such as 3D climbing surface) is generally used for profile cutting. However, this method has the problems of low machining efficiency and long time consumption, and due to the constantly changing contact points of the tool and the curved surface, it is easy to cause unstable machining surface quality and poor finish, which is difficult to meet the increasingly stringent industrial requirements.

[0003] Therefore, a geometric parameter collaborative optimization method of a taper hole machining tool is proposed. SUMMARY

[0004] The present application aims to provide a geometric parameter collaborative optimization method of a taper hole machining tool, which constructs a high-fidelity digital twin model, integrates multi-physics field coupling and machine tool dynamic characteristics based on the finite element method, generates a geometric parameter-performance data mapping set and trains a computational proxy model; outputs a Pareto solution set through multi-objective global optimization; constructs a constraint range based on the solution set, calls the digital twin model for local optimization, and obtains the optimal geometric parameter combination. The method includes a self-correcting mechanism that corrects the material constitutive relationship and friction coefficient through experimental data; uses stage-by-stage adaptive learning, NSGA-II and DBSCAN clustering, and the accompanying method to optimize efficiency; introduces the Pareto stability index and transfer learning to improve the robustness and cross-task reusability of the results. The present application can realize the collaborative optimization of the geometric parameters of a high-precision and high-efficiency taper hole machining tool.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] A geometric parameter collaborative optimization method of a taper hole machining tool, comprising:

[0007] S1. Construct a high-fidelity digital twin model with the taper hole machining tool geometric parameters as input. The model is based on the finite element method, integrates the tool-workpiece interaction model of cutting force-cutting heat-deformation multi-physics field coupling and the dynamic characteristic model of the machine tool-clamp-tool-workpiece system, and outputs the predicted values of size accuracy, surface roughness and machining efficiency corresponding to the geometric parameters through multi-physics field coupling simulation, to generate a geometric parameter-performance data mapping set; based on the mapping set as training samples, a computational proxy model is trained;

[0008] S2. Perform multi-objective global optimization based on the computational proxy model and a preset multi-element performance objective function, and output a candidate Pareto optimal solution set; the preset multi-element performance objective function is constructed based on machining efficiency, surface roughness and size accuracy.

[0009] S3. Selecting an initial candidate solution set from a candidate Pareto optimal solution set, and constructing a geometric parameter constraint range based on the initial candidate solution set; calling the digital twin model based on the geometric parameter constraint range, and constructing a preset single-objective function for local optimization search to obtain an optimal geometric parameter combination; the single-objective function is a comprehensive evaluation function constructed based on surface roughness and dimensional accuracy.

[0010] Preferably, the geometric parameters at least include a tool nose radius, an opening R angle, a rake angle and a relief angle; and the performance indicators at least include dimensional accuracy, surface roughness and machining efficiency.

[0011] Preferably, the high-fidelity digital twin model further comprises a self-correction mechanism, specifically comprising:

[0012] Manufacturing a physical tool based on the geometric parameters, and obtaining real physical data in a cutting experiment, including but not limited to cutting force, temperature and workpiece deformation;

[0013] Comparing the real physical data with the simulation prediction value of the high-fidelity digital twin model, and calculating a deviation;

[0014] Using a system identification algorithm to automatically correct physical parameters in the high-fidelity digital twin model, including but not limited to material constitutive relation and friction coefficient, according to the deviation;

[0015] Updating the corrected model parameters to the geometric parameter-performance data mapping set for training a computational proxy model.

[0016] Preferably, the training process of the computational proxy model is a phased adaptive learning, specifically comprising:

[0017] Global exploration stage: based on an initial sample point combination, a method based on uncertainty sampling is used for preliminary training, taking maximizing reduction of global prediction uncertainty of the model as a criterion, new sample points are iteratively selected and the high-fidelity digital twin model is called for simulation, until the uncertainty reduction rate is lower than a first threshold set based on a computational resource constraint;

[0018] Local utilization stage: automatically switching to this stage, taking maximum expected improvement as a criterion, new sample points are iteratively selected and simulated in the current excellent area, until the expected improvement is lower than a second threshold set based on an accuracy requirement, forming a final computational proxy model; the proxy model is constructed based on a neural network;

[0019] Preferably, the specific acquisition process of the candidate Pareto optimal solution set is:

[0020] The second generation non-dominated sorting genetic algorithm is adopted to perform global multi-objective optimization on the computing agent model based on the optimized population size and iteration number of the computing resource, to generate a preliminary Pareto solution set;

[0021] The DBSCAN clustering algorithm based on density is adopted on the preliminary Pareto solution set, to identify and output multiple separated and high-density performance clusters as the input of the initial candidate solution set by adaptively adjusting the density radius and the minimum number of points.

[0022] Preferably, the construction process of the geometric parameter constraint range comprises:

[0023] Statistical analysis is performed on the initial candidate solution set to extract the distribution range of each geometric parameter, including but not limited to mean, variance and boundary value;

[0024] Based on the distribution range, the feasible constraint range of the geometric parameter is generated in combination with the manufacturing tolerance and the processing equipment constraint;

[0025] Within the constraint range, the high-fidelity digital twin model is called for simulation verification.

[0026] Preferably, the local optimization search adopts a gradient-based optimization method, specifically including:

[0027] The centroid point of the performance cluster is selected from the initial candidate solution set as the initial solution, and the high-fidelity digital twin model is called for simulation;

[0028] The real physical gradient of performance to geometric parameter is calculated by the finite element adjoint solver through the adjoint method of the digital twin model;

[0029] Based on the physical gradient, local optimization is performed to generate a physical optimal solution as the optimal geometric parameter combination.

[0030] Preferably, the method further comprises an optimization post-processing and knowledge management step, specifically including:

[0031] The Pareto stability index is introduced to couple the performance score calculated based on the multi-criteria decision method and the performance fluctuation analysis result within the manufacturing tolerance range through a preset function, and output the geometric parameter combination with the highest PSI index;

[0032] The knowledge asset formed after the completion of the optimization task, containing all high-fidelity sample points and the final computing agent model, is structured and archived as a reusable knowledge base;

[0033] When a new workpiece material and similar processing requirements are executed, the knowledge asset is loaded as a pre-trained model and prior knowledge base to initialize the optimization process of the new task;

[0034] By comparing the cutting experiment data of the physical cutter with the predicted values of the high-fidelity digital twin model, the model deviation is analyzed and the performance prediction of the subsequent task is optimized.

[0035] Compared with the prior art, the beneficial effects of the present application are:

[0036] 1. The present application provides a double-mode driving optimization method combining a high-fidelity digital twin model and a computational agent model. In the global optimization stage, the present application uses a high-fidelity digital twin model to generate an initial mapping set through finite element physical simulation; a computational agent model is trained based on this data set, and NSGA-II is used to generate a candidate Pareto optimal solution set of potential optimal solutions, overcoming the dependence on finite element simulation for global optimization in the prior art; in the local optimization stage, the present application locks a high-potential performance cluster through the DBSCAN clustering algorithm, and switches the optimization subject back to the high-fidelity digital twin model to ensure the physical accuracy of the final solution; further, the present application innovatively uses the adjoint method for local optimization search in this stage; the true physical gradient of the objective function with respect to all design variables can be obtained through finite element adjoint solver calculation, and the efficiency is improved by orders of magnitude compared with the traditional gradient calculation method based on perturbation; the efficient gradient optimization method is combined with the high-precision physical model to ensure the accuracy of the local optimization process; in summary, the present application realizes the organic unification of computational efficiency and solution accuracy through the cooperative strategy of "agent model global exploration" and "high-fidelity model local accurate optimization", and solves the contradiction between optimization accuracy and globality in the traditional geometric parameter cooperative optimization method of taper hole machining tools.

[0037] 2、The closed-loop optimization system closely combines virtual simulation and physical experiment; firstly, the self-correction mechanism contained in the application obtains real physical data such as cutting force and temperature by manufacturing physical cutters and performing cutting experiments; by comparing the experimental data with the simulation prediction value of the high-fidelity digital twin model, the physical parameters in the model, such as the material constitutive relation (Johnson-Cook model parameters) and the friction coefficient, are automatically corrected according to the deviation by using the system identification algorithm (sequential quadratic programming SQP); the closed-loop feedback and dynamic correction process ensures the long-term high fidelity of the digital twin model, enables it to continuously self-improve against the physical reality, fundamentally guarantees the reliability of its prediction, provides a high-credibility data basis for the entire optimization process, and further improves the geometric parameter collaborative optimization efficiency of the taper hole machining tool; and the Pareto stability index is innovatively introduced in the optimization post-processing link; a multi-dimensional evaluation system is constructed through the index, and the theoretical performance score calculated based on the multi-criteria decision method (such as TOPSIS) is coupled with the performance fluctuation analysis results based on the manufacturing tolerance range; the performance fluctuation analysis quantifies the sensitivity of the candidate solution to manufacturing errors; solves the problem that the Pareto optimal solution in the prior art often only pursues theoretical performance, while ignoring the sharp decline in performance that may be caused by parameter fluctuation in the actual manufacturing process, and significantly improves the processing quality.

[0038] 3、The application designs a phased adaptive learning strategy to guide the training of the computational agent model in view of the huge consumption of high-fidelity simulation computing resources; in the global exploration stage, the high-fidelity model is preferentially called for simulation in the information sparse area to maximize the reduction of the global prediction uncertainty of the model; after the uncertainty drops to a preset threshold, the system automatically switches to the local utilization stage, and maximizes the expected improvement (EI) in the current excellent area for fine sampling; this intelligent active learning strategy ensures that each expensive simulation calculation serves the most urgent information demand, realizes the goal of constructing a high-precision agent model with the least simulation times, and significantly reduces the computing resource consumption and time cost; and a knowledge management and transfer learning mechanism is established; after each optimization task is completed, the knowledge assets of all high-fidelity sample points and the final computational agent model are structured and archived to form a reusable knowledge base, thereby providing a data basis for further improving the processing quality. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 A geometric parameter collaborative optimization method for a taper hole machining tool provided by the embodiment of the application;

[0040] Figure 2 A geometric parameter collaborative optimization method for a taper hole machining tool provided by the embodiment of the application; DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0042] Embodiment one

[0043] In a plasma etching device for semiconductor manufacturing, a high-purity quartz ring is a key component for protecting and fixing a wafer. The component is distributed with high-precision taper holes for gas flow guiding or fixing; since the quartz material is hard and brittle, and the surface finish and dimensional accuracy after processing are extremely high, in order to improve the processing quality and efficiency of the taper hole of the high-purity quartz ring, a geometric parameter collaborative optimization method of taper hole processing tool is applied; such as Figure 1 A geometric parameter collaborative optimization method of taper hole processing tool provided for the embodiments of the present application is shown in the step schematic diagram;

[0044] A geometric parameter collaborative optimization method of taper hole processing tool, comprising:

[0045] S1. Construct a high-fidelity digital twin model, taking the geometric parameters of the taper hole processing tool as input, the model is based on the finite element method, integrating the tool-workpiece interaction model of multi-physical field coupling of cutting force-cutting heat-deformation and the dynamic characteristic model of the machine tool-clamp-tool-workpiece system, through multi-physical field coupling solving simulation, output the predicted values of size accuracy, surface roughness and processing efficiency corresponding to the geometric parameters, generate a geometric parameter-performance data mapping set; based on the mapping set as a training sample, a calculation agent model is trained;

[0046] The high-fidelity digital twin model is realized by the Coupled Field Analysis module of ANSYS Mechanical 2023 R1, and the cutting force, heat and deformation are solved by thermal-structural bidirectional coupling. The tool uses tetrahedral mesh, the mesh size is 0.01 mm, the workpiece mesh size is 0.05-0.5 mm (the mesh size near the tool tip is encrypted to 0.02 mm), and the total number of meshes is about 500,000. The boundary conditions include the workpiece bottom surface fixed, the tool applied with a feed rate of 0.1 mm / rev, a cutting depth of 0.05 mm, and a rotation speed of 8,000 rpm. The thermal boundary condition is an ambient temperature of 25°C, and the tool convective heat transfer coefficient is 20 W / (m²·K). The solving setting is 5 iterations per step, and the convergence residual is <10⁻ 5, time step 0.01 s. Mesh convergence analysis shows that the tool mesh is 0.01 mm, Changes <1.5%, Ra changes <2%, to ensure simulation accuracy.

[0047] Further, the geometric parameters at least include tool nose radius, opening R angle, rake angle and relief angle; the performance indicators at least include dimensional accuracy, surface roughness and machining efficiency.

[0048] Further, the high-fidelity digital twin model is based on the finite element method and is realized by ANSYS Mechanical, integrating a tool-workpiece interaction model and a dynamic characteristic model of a machine tool-clamp-tool-workpiece system; the tool-workpiece interaction model is used to simulate the multi-physical field coupling between the tool and the workpiece during the machining process, including cutting force, cutting heat and workpiece deformation;

[0049] Cutting force: the stress-strain response of quartz material under the action of tool cutting is calculated by Johnson-Cook constitutive model, so as to determine the size of cutting force, which directly affects the final dimensional accuracy and surface quality; Johnson-Cook parameters (A, B, n, C, m) are A=150 MPa, B=200 MPa, n=0.3, C=0.01, m=1.0;

[0050] Cutting heat: the model is based on heat conduction equation and friction model, simulating the heat generated in the contact area between tool and workpiece due to friction and plastic deformation, and predicting the temperature distribution;

[0051] Workpiece deformation: the deformation of the workpiece under the action of cutting force is calculated by linear elastic equation and maximum principal stress criterion (set the fracture stress of quartz to 50 MPa).

[0052] The tool-workpiece interaction model adopts Johnson-Cook constitutive model, with parameters A=150 MPa, B=200 MPa, n=0.3, C=0.01, m=1.0, which is suitable for hard and brittle quartz (elastic modulus 72 GPa, Poisson's ratio 0.17, thermal conductivity 1.4 W / (m·K), fracture stress 50 MPa). The cutting heat is based on the heat conduction equation, and the boundary conditions include ambient temperature 25°C and tool convective heat transfer coefficient 20 W / (m²·K). The friction model adopts Coulomb friction, with friction coefficient μ=0.2, and contact pressure is calculated by Hertz contact theory;

[0053] Dynamic characteristic model of machine tool-clamp-tool-workpiece system: this model is used to simulate the vibration characteristics of the machine tool during the machining process, such as machine tool stiffness (50 N / μm), damping ratio (0.05) and natural frequency (about 500 Hz);

[0054] The main natural frequency 500 Hz and the secondary frequency 800 Hz are extracted by the ANSYS Modal module, and the error is verified to be less than 5%.

[0055] Further, The formula is:

[0056] ;

[0057] ;

[0058] Among them, , ; ; is the vibration amplitude;

[0059] Multi-physical field coupling solution and output: the above two models are integrated by the ANSYS Coupled Field Analysis module to realize comprehensive coupling simulation of cutting force, heat, deformation and vibration.

[0060] Specifically, the above two models are integrated by the ANSYS Coupled Field Analysis module to realize comprehensive coupling simulation of cutting force, heat, deformation and vibration. Specifically, the integration adopts a one-way coupling mode, and in each simulation time step, the real-time vibration displacement of the tool calculated by the dynamic characteristic model (obtained by ANSYS Modal module analysis) is superimposed on the preset feed motion of the tool in the multi-physical field coupling model as an additional boundary condition;

[0061] The simulation finally outputs the performance prediction value corresponding to each group of geometric parameters, specifically including dimensional accuracy (D, μm), surface roughness (Ra, μm) and machining efficiency (V, mm³ / s);

[0062] Mapping set generation: 500 groups of geometric parameter combinations (r, R, γ, α) are generated using Latin Hypercube Sampling (LHS); first, according to the conventional tool design experience of machining hard and brittle quartz materials, the initial optimization range of geometric parameters is set; for example, in this embodiment, it is set that: the tool tip roundness r∈[0.1mm, 0.8mm], the opening angle R∈[0.2mm, 1.0mm], the rake angle γ∈[5°, 15°], and the relief angle α∈[8°, 18°]; and through ANSYS simulation, the corresponding performance (D, Ra, V) is output to form a mapping set.

[0063] ​​​​​​r represents a tool tip fillet; R represents an opening R angle; γ represents a rake angle; a represents a relief angle; represents dimensional accuracy; represents surface roughness; represents machining efficiency;

[0064] Further, the high-fidelity digital twin model further comprises a self-correction mechanism, specifically comprising:

[0065] manufacturing a physical tool based on the geometric parameters, obtaining real physical data in a cutting experiment, including but not limited to cutting force, temperature and workpiece deformation;

[0066] comparing the real physical data with the simulation prediction value of the high-fidelity digital twin model, and calculating the deviation;

[0067] using a system identification algorithm to automatically correct the physical parameters in the high-fidelity digital twin model, including but not limited to material constitutive relation and friction coefficient; for example, adjusting Johnson-Cook parameters;

[0068] The system identification algorithm is specifically a parameter optimization algorithm based on sequential quadratic programming (SQP). The algorithm constructs a function with the objective of minimizing the error between simulation and experimental data, specifically:

[0069] ;

[0070] wherein, and are the simulation cutting force and temperature; and are the corresponding experimental measurement values; and are preset weight coefficients; the algorithm takes the parameters A, B, n, C, m of the material constitutive model (Johnson-Cook model) as design variables, and minimizes the objective function through iterative optimization, thereby realizing automatic and accurate correction of the key physical parameters of the model.

[0071] Further, the specific self-correction mechanism comprises:

[0072] Physical data acquisition step: first, based on the preset geometric parameters, a corresponding physical tool is manufactured; the physical tool is used for cutting experiment, and real physical data in the experimental process is collected through a sensor. The real physical data includes but is not limited to cutting force, machining area temperature and workpiece deformation;

[0073] Bias calculation step: comparing the real physical data with the simulation prediction value output by the high-fidelity digital twin model based on the same geometric parameters and working condition; obtaining the bias between the real physical data and the simulation prediction value.

[0074] Model parameter correction step: calling system identification algorithm, automatically correcting and adjusting the physical parameters in the high-fidelity digital twin model according to the bias; the physical parameters include but are not limited to material constitutive relation for describing material stress-strain response (such as adjusting Johnson-Cook model parameters) and friction coefficient of tool-workpiece contact surface.

[0075] Model and data updating step: updating the corrected physical parameters to the high-fidelity digital twin model, thereby obtaining the optimized model. Finally, calling the optimized model, re-updating the geometric parameter-performance data mapping set, and using the mapping set as the input sample for training the computational surrogate model to ensure the accuracy and reliability of the subsequent optimization process.

[0076] Updating the corrected model parameters to the geometric parameter-performance data mapping set for training the computational surrogate model.

[0077] Further, the training process of the computational surrogate model is a phased adaptive learning, specifically including:

[0078] Global exploration stage: based on the initial sample point combination, using the method based on uncertainty sampling for preliminary training, taking the maximum reduction of global prediction uncertainty as the criterion, iteratively selecting new sample points and calling the high-fidelity digital twin model for simulation, until the uncertainty reduction rate is lower than the first threshold set based on the computational resource constraint;

[0079] Local utilization stage: automatically switched to this stage, taking the maximum expected improvement as the criterion, iteratively selecting new sample points and performing simulation in the current excellent area, until the expected improvement is lower than the second threshold set based on the accuracy requirement, forming the final computational surrogate model; the surrogate model is based on neural network construction;

[0080] The neural network is a fully connected feedforward neural network, and its specific architecture and training parameter settings are as follows:

[0081] Network architecture: containing 1 input layer, 2 hidden layers and 1 output layer. The input layer is provided with 4 neurons, respectively corresponding to the four geometric parameters of tool nose radius (r), opening R angle (R), rake angle (γ) and relief angle (α). Each hidden layer is provided with 32 neurons, and the rectified linear unit (ReLU) is uniformly used as the activation function to enhance the nonlinear fitting ability of the model. The output layer is provided with 3 neurons, respectively corresponding to the dimensional accuracy (δ), surface roughness (Ra) and surface quality (Q) of the workpiece. surface roughness (Ra) ) and machining efficiency (Machining Efficiency, ME) ) three performance indicators.

[0082] The first threshold value specifically refers to that in the continuous 3 iterations, the relative decline rate of the model global prediction uncertainty (measured by the mean of the Gaussian process regression prediction variance) is less than 5%. The second threshold value specifically refers to that the expected improvement (EI) value is less than 1% of the target precision requirement of the main performance indicator (such as surface roughness (Ra) ) For example, if Ra is required to reach 0.2 μm, when the EI value is less than 0.002 μm, the local utilization stage stops.

[0083] The expected improvement (EI) criterion is as follows:

[0084] ;

[0085] wherein, is a set of input geometric parameters; is the current optimal value; and are the mean and standard deviation predicted by the Gaussian process regression model at point ; and are the cumulative distribution function and probability density function of the standard normal distribution, respectively;

[0086] The self-correction mechanism verifies the prediction accuracy of the high-fidelity digital twin model through experiments, corrects the key physical parameters (such as material constitutive relation, friction coefficient), ensures that the model is consistent with the actual machining process, and improves the reliability of the geometric parameter-performance data mapping set, providing high-quality data for training the computational surrogate model.

[0087] S2. Perform multi-objective global optimization based on the computational surrogate model and the preset multi-performance objective function, and output a candidate Pareto optimal solution set; the preset multi-performance objective function is constructed based on machining efficiency, surface roughness and dimensional accuracy;

[0088] The multi-performance objective function is as follows:

[0089] ;

[0090] wherein, denotes the multi-performance objective function; denotes a minimization function; this function aims to minimize the surface roughness (Ra) , minimize the dimensional accuracy (D ) and maximize the machining efficiency (Machining Efficiency, ME) ;

[0091] S3. Select an initial candidate solution set from a candidate Pareto optimal solution set, and construct a geometric parameter constraint range based on the initial candidate solution set; call the digital twin model based on the geometric parameter constraint range, and construct a preset single-objective function for local optimization search to obtain an optimal geometric parameter combination; the single-objective function is a comprehensive evaluation function constructed based on surface roughness and dimensional accuracy.

[0092] The single-objective function is:

[0093]

[0094] wherein, represents the single-objective function, , is a weight coefficient; is a minimum value of dimensional accuracy; is a maximum value of dimensional accuracy; is a minimum value of surface roughness; is a maximum value of surface roughness; , are respectively 0.5 and 0.5;

[0095] Further, the specific acquisition process of the candidate Pareto optimal solution set is as follows:

[0096] A second-generation non-dominated sorting genetic algorithm is adopted to perform global multi-objective optimization on the computing agent model based on the population size and the number of iterations of computing resource optimization, to generate a preliminary Pareto solution set;

[0097] A DBSCAN clustering algorithm based on density is adopted on the preliminary Pareto solution set, to identify and output multiple separated and high-density performance clusters as the input of the initial candidate solution set, through adaptive adjustment of the density radius and the minimum number of points.

[0098] Multiple separated and high-density performance clusters are identified and output through adaptive adjustment of the density radius and the minimum number of points. The adaptive adjustment method is as follows: the distances (k-dist, k=4) of each solution to its fourth nearest neighbor point are calculated and sorted, and the corresponding distance value at the 'inflection point' of the k-distance graph is automatically selected as the density radius, and the minimum number of points is set to 4.

[0099] Further, the construction process of the geometric parameter constraint range includes:

[0100] Statistical analysis is performed on the initial candidate solution set, to extract the distribution range of each geometric parameter, including but not limited to the mean, variance and boundary value;

[0101] Based on the distribution range, the feasible constraint range of the geometric parameter is generated in combination with the manufacturing tolerance and the processing equipment constraint; ​

[0102] Within the constraint range, the high-fidelity digital twin model is called for simulation verification.

[0103] Further, the local optimization search adopts a gradient-based optimization method, specifically including:

[0104] The centroid of the performance cluster is selected from the initial candidate solution set as the initial solution, and the high-fidelity digital twin model is called for simulation;

[0105] The real physical gradient of performance to geometric parameters is calculated by a finite element adjoint solver using the adjoint method of the digital twin model;

[0106] Based on the physical gradient, local optimization is performed to generate a physical optimal solution as the optimal geometric parameter combination.

[0107] Specifically, in ANSYS, the single-objective function is taken as the optimization target, and the geometric parameters (r, R, γ, α) are defined as design variables; the 'Sensitivity Analysis' function in the optimization module of ANSYS is called. This function is solved in the background using the adjoint method to calculate the real physical gradient of to all design variables;

[0108] Further, the method further includes an optimization post-processing and knowledge management step, specifically including:

[0109] The Pareto stability index is introduced to evaluate the robustness of the candidate solution under the influence of manufacturing tolerances, and the performance score calculated based on the multi-criteria decision method is coupled with the performance fluctuation analysis result within the manufacturing tolerance range to output the geometric parameter combination with the highest PSI index;

[0110] ;

[0111] wherein, Pareto stability index (PSI) is represented; MCDM score is represented; , preset weight coefficient is represented; ; performance stability within the manufacturing tolerance (±0.01 mm) is represented;

[0112] The multi-criteria decision method score calculates the performance score of each candidate solution by TOPSIS (Technique for Order Preference by Similarity to Ideal Solution); the specific process is as follows:

[0113] Based on the candidate Pareto solution set, a decision matrix is constructed, in which each candidate solution is a row and the three performance indicators (dimensional accuracy, surface roughness and machining efficiency) are columns.

[0114] The positive ideal solution (PIS) and the negative ideal solution (NIS) are determined. The PIS is composed of the optimal values of each indicator; the NIS is composed of the worst values of each indicator;

[0115] The Euclidean distance of each candidate solution to the PIS and NIS is calculated;

[0116] According to the distance, the relative closeness of each solution is calculated, which is the multi-criteria decision-making (MCDM) score, whose value is between 0 and 1, and the closer to 1, the better the comprehensive performance;

[0117] The performance fluctuation analysis result is used to quantify the severity of the performance change caused by the fluctuation of the geometric parameters within the manufacturing tolerance range, representing the robustness of the solution; the specific process is as follows:

[0118] For a candidate solution of geometric parameter combination (such as tool nose radius r, opening R angle,...), within the range of its respective manufacturing tolerance (for example, ±0.01mm), N (for example, N=100) random parameter samples are generated by Monte Carlo or Latin hypercube sampling;

[0119] Input the N samples into the trained computational agent model to quickly predict N sets of corresponding performance indicators;

[0120] Calculate the coefficient of variation (CV = standard deviation / mean) of each performance indicator; that is, and Weighted;

[0121] The performance stability is defined as the reciprocal of the weighted sum of the coefficients of variation; specifically, ; represents the weighted sum of the coefficients of variation;

[0122] The knowledge asset containing all high-fidelity sample points and the final computational agent model formed after the completion of the optimization task is structured and archived as a reusable knowledge base;

[0123] When performing a new workpiece material and similar machining requirement optimization task, load the knowledge asset as a pre-trained model and prior knowledge base to initialize the optimization process of the new task;

[0124] By comparing the experimental data of the physical tool with the predicted values of the high-fidelity digital twin model, the model deviation is analyzed and the performance prediction of the subsequent task is optimized.

[0125] In summary, as Figure 2A geometric parameter cooperative optimization method of a taper hole machining tool provided by the embodiment of the application is shown in the whole structure schematic diagram; the embodiment generates a geometric parameter-performance data mapping set and trains a calculation agent model by constructing a high-fidelity digital twin model, integrating multi-physical field coupling and machine tool dynamic characteristics based on the finite element method; a Pareto solution set is output through multi-objective global optimization; a constraint range is constructed based on the solution set, and a digital twin model is called for local optimization to obtain an optimal geometric parameter combination. The method includes a self-correcting mechanism, which corrects the material constitutive relation and the friction coefficient through experimental data; a phased adaptive learning, NSGA-II and DBSCAN clustering, and an accompanying method are used to optimize efficiency; a Pareto stability index and migration learning are introduced to improve the robustness and cross-task reusability of the results. The application can realize high-precision and high-efficiency geometric parameter cooperative optimization of the taper hole machining tool.

[0126] The embodiment provides a double-mode driven optimization method combining a high-fidelity digital twin model and a calculation agent model. In the global optimization stage, the application generates an initial mapping set through finite element physical simulation by using a high-fidelity digital twin model; a calculation agent model is trained based on the data set, and NSGA-II is used to generate a candidate Pareto optimal solution set of potential optimal solutions to overcome the global optimization of the prior art which simply relies on finite element simulation; in the local optimization stage, the application locks a high-potential performance cluster through a DBSCAN clustering algorithm, and switches the optimization subject back to the high-fidelity digital twin model to ensure the physical accuracy of the final solution; further, the application innovatively uses an accompanying method for local optimization search in this stage; the real physical gradient of the objective function to all design variables can be obtained through finite element accompanying solver calculation, and compared with the traditional gradient calculation method based on perturbation, the efficiency is improved by orders of magnitude; the efficient gradient optimization method is combined with the high-precision physical model to ensure the accuracy of the local optimization process; in summary, the application realizes the organic unification of calculation efficiency and solution accuracy through the cooperative strategy of "agent model global exploration" and "high-fidelity model local accurate optimization", and solves the contradiction between optimization accuracy and globality in the traditional geometric parameter cooperative optimization method of the taper hole machining tool.

[0127] The embodiment provides a closed-loop optimization system combining virtual simulation and physical experiment closely; firstly, the self-correction mechanism contained in the application obtains real physical data such as cutting force and temperature by manufacturing physical cutters and performing cutting experiments; by comparing the experimental data with the simulation prediction value of the high-fidelity digital twin model, the physical parameters in the model, such as the material constitutive relation (Johnson-Cook model parameters) and the friction coefficient, are automatically corrected according to the deviation by using a system identification algorithm (sequential quadratic programming SQP); the closed-loop feedback and dynamic correction process ensures the long-term high fidelity of the digital twin model, enables it to continuously self-improve against physical reality, fundamentally guarantees the reliability of its prediction, provides a high-credibility data basis for the entire optimization process, and further improves the geometric parameter collaborative optimization efficiency of the taper hole machining cutter; and the Pareto stability index is innovatively introduced in the optimization post-processing link; a multi-dimensional evaluation system is constructed through the index, and the theoretical performance score calculated based on the multi-criteria decision method (such as TOPSIS) is coupled with the performance fluctuation analysis result based on the manufacturing tolerance range through a preset function; the performance fluctuation analysis quantifies the sensitivity of the candidate solution to manufacturing errors; solves the problem that the Pareto optimal solution in the prior art often only pursues theoretical performance, and ignores the sharp decline in performance that may be caused by parameter fluctuation in the actual manufacturing process, and significantly improves the processing quality.

[0128] The embodiment designs a phased adaptive learning strategy to guide the training of the computational agent model in view of the problem of huge consumption of high-fidelity simulation computing resources. In the global exploration stage, the high-fidelity model is called for simulation in the information sparse area as the criterion for maximizing the reduction of the global prediction uncertainty of the model; after the uncertainty is reduced to a preset threshold, the local utilization stage is automatically switched to, and the criterion for maximizing the expected improvement (EI) is used to perform fine sampling in the current excellent area. This intelligent active learning strategy ensures that each expensive simulation calculation serves the most urgent information demand, realizes the goal of constructing a high-precision agent model with the least simulation times, and significantly reduces the computing resource consumption and time cost; and a knowledge management and transfer learning mechanism is established; after each optimization task is completed, the knowledge assets of all high-fidelity sample points and the final computational agent model are structured and archived to form a reusable knowledge base, thereby providing a data basis for further improving the processing quality.

[0129] Embodiment two

[0130] In the plasma etching equipment of semiconductor manufacturing, the high-purity quartz ring is a key component for protecting and fixing the wafer. The component is distributed with high-precision taper holes for gas guiding or fixing; since the quartz material is hard and brittle, and the surface finish and dimensional accuracy after processing are extremely high, in order to improve the processing quality and efficiency of the taper hole of the high-purity quartz ring, a geometric parameter collaborative optimization method of taper hole processing tool is applied; such as Figure 1 A geometric parameter collaborative optimization method of taper hole processing tool provided for an embodiment of the application is shown in a step schematic diagram;

[0131] A geometric parameter collaborative optimization method of taper hole processing tool comprises:

[0132] S1. A high-fidelity digital twin model is constructed, taking the geometric parameters of the taper hole processing tool as input, the model is based on the finite element method, integrates the tool-workpiece interaction model of multi-physical field coupling of cutting force-cutting heat-deformation and the dynamic characteristic model of the machine tool-clamp-tool-workpiece system, and outputs the predicted values of size accuracy, surface roughness and processing efficiency corresponding to the geometric parameters through multi-physical field coupling simulation, to generate a geometric parameter-performance data mapping set; based on the mapping set as a training sample, a calculation agent model is trained;

[0133] S2. Based on the calculation agent model and the preset multi-element performance objective function, multi-objective global optimization is performed, and a candidate Pareto optimal solution set is output; the preset multi-element performance objective function is constructed based on processing efficiency, surface roughness and dimensional accuracy;

[0134] S3. Select an initial candidate solution set from the candidate Pareto optimal solution set, and construct a geometric parameter constraint range based on the initial candidate solution set; call the digital twin model based on the geometric parameter constraint range, and construct a preset single objective function for local optimization search to obtain an optimal geometric parameter combination; the single objective function is a comprehensive evaluation function constructed based on surface roughness and dimensional accuracy.

[0135] Further, the geometric parameters at least include tool nose radius, opening R angle, rake angle and relief angle; and the performance indicators at least include dimensional accuracy, surface roughness and processing efficiency.

[0136] Further, the high-fidelity digital twin model further comprises a self-correction mechanism, specifically comprising:

[0137] Manufacture the physical tool based on the geometric parameters, and obtain the real physical data in the cutting experiment, including but not limited to cutting force, temperature and workpiece deformation;

[0138] Compare the real physical data with the simulation predicted value of the high-fidelity digital twin model, and calculate the deviation;

[0139] Utilizing system identification algorithms, automatically correct physical parameters in the high-fidelity digital twin model, including but not limited to material constitutive relations and friction coefficients, based on deviations;

[0140] Update the corrected model parameters to the set of geometry-parameter-performance data mappings for training of the computational surrogate model.

[0141] Further, the training process of the computational surrogate model is a phased adaptive learning, specifically including:

[0142] Global exploration phase: based on the initial sample point combination, a method based on uncertainty sampling is used for preliminary training, taking maximizing the reduction of global prediction uncertainty of the model as the criterion, iteratively selecting new sample points and calling the high-fidelity digital twin model for simulation, until the uncertainty reduction rate is lower than the first threshold set based on the computational resource constraint;

[0143] Local utilization phase: automatically switch to this phase, taking the maximum expected improvement as the criterion, iteratively selecting new sample points and performing simulation in the current superior region, until the expected improvement is lower than the second threshold set based on the accuracy requirement, forming the final computational surrogate model; the surrogate model is constructed based on neural networks;

[0144] Further, the specific acquisition process of the candidate Pareto optimal solution set is:

[0145] A second-generation non-dominated sorting genetic algorithm is used to perform global multi-objective optimization on the computational surrogate model based on the population size and iteration number optimized for computational resources, generating a preliminary Pareto solution set;

[0146] A density-based DBSCAN clustering algorithm is used on the preliminary Pareto solution set to identify and output multiple separate, high-density performance clusters as inputs for the initial candidate solution set through adaptive adjustment of the density radius and minimum number of points.

[0147] Further, the construction process of the geometry parameter constraint range includes:

[0148] Statistical analysis is performed on the initial candidate solution set to extract the distribution range of each geometry parameter, including but not limited to mean, variance, and boundary value;

[0149] Based on the distribution range, combined with manufacturing tolerances and processing equipment constraints, generate feasible constraint ranges for geometry parameters;

[0150] Within the constraint range, call the high-fidelity digital twin model for simulation verification.

[0151] Further, the local optimization search uses a gradient-based optimization method, specifically including:

[0152] select a performance cluster centroid point from the initial candidate solution set as an initial solution, and call the high-fidelity digital twin model to perform simulation;

[0153] using the adjoint method of the digital twin model, calculating the real physical gradient of performance to geometric parameters through a finite element adjoint solver;

[0154] based on the physical gradient, performing local optimization to generate a physical optimal solution as the optimal geometric parameter combination.

[0155] Further, the method further comprises an optimization post-processing and knowledge management step, specifically comprising:

[0156] introducing a Pareto stability index, coupling the performance score calculated based on the multi-criteria decision method and the performance fluctuation analysis result within the manufacturing tolerance range through a preset function, and outputting the geometric parameter combination with the highest PSI index;

[0157] structuring and archiving the knowledge assets formed after the completion of the optimization task, including all high-fidelity sample points and the final calculation agent model, as a reusable knowledge base;

[0158] when a new workpiece material and similar machining requirements are executed to perform an optimization task, loading the knowledge assets as a pre-training model and prior knowledge base to initialize the optimization process of the new task;

[0159] using the comparison between the entity tool cutting experimental data and the high-fidelity digital twin model prediction value to analyze the model deviation and optimize the performance prediction of subsequent tasks.

[0160] In summary, as Figure 2 A geometric parameter collaborative optimization method of a taper hole machining tool provided by the embodiment of the present application is shown in the whole structure schematic view. The embodiment constructs a high-fidelity digital twin model, integrates multi-physical field coupling and machine tool dynamic characteristics based on the finite element method, generates a geometric parameter-performance data mapping set and trains a calculation agent model; outputs a Pareto solution set through multi-objective global optimization; constructs a constraint range based on the solution set, calls the digital twin model to perform local optimization, and obtains an optimal geometric parameter combination. The method includes a self-correction mechanism, which corrects the material constitutive relation and the friction coefficient through experimental data; adopts phased adaptive learning, NSGA-II and DBSCAN clustering, and adjoint method to optimize efficiency; introduces a Pareto stability index and migration learning to improve the robustness and cross-task reusability of the results. The present application can realize high-precision and high-efficiency geometric parameter collaborative optimization of a taper hole machining tool.

[0161] This embodiment provides a dual-mode driven optimization method combining a high-fidelity digital twin model and a computational proxy model. In the global optimization stage, the invention utilizes the high-fidelity digital twin model to generate an initial mapping set through finite element physical simulation; based on this dataset, a computational proxy model is trained, and NSGA-II is used to generate a candidate Pareto optimal solution set for potential optimal solutions, overcoming the limitations of existing technologies that rely solely on finite element simulation for global optimization. In the local optimization stage, the invention uses the DBSCAN clustering algorithm to identify high-potential performance clusters and switches the optimization focus back to the high-fidelity digital twin model to ensure the physical accuracy of the final solution. Furthermore, the invention innovatively employs the adjoint method for local optimization in this stage. The invention employs a superior search method, capable of calculating the true physical gradient of the objective function with respect to all design variables using a finite element adjoint solver. Compared to traditional perturbation-based gradient calculation methods, its efficiency is improved by orders of magnitude. By combining a high-efficiency gradient optimization method with a high-precision physical model, the accuracy of the local optimization process is guaranteed. In summary, this invention achieves an organic balance between computational efficiency and solution accuracy through a collaborative strategy of "global exploration of the surrogate model" and "local accurate optimization of the high-fidelity model," resolving the contradiction between optimization accuracy and globality in traditional methods for co-optimizing geometric parameters of tapered hole machining tools.

[0162] This embodiment provides a closed-loop optimization system that closely integrates virtual simulation with physical experiments. First, the self-calibration mechanism included in this invention obtains real physical data such as cutting force and temperature by manufacturing physical cutting tools and conducting cutting experiments. By comparing the experimental data with the simulation predictions of a high-fidelity digital twin model, a system identification algorithm (Sequential Quadratic Programming, SQP) is used to automatically correct physical parameters in the model based on deviations, such as material constitutive relations (Johnson-Cook model parameters) and friction coefficients. This closed-loop feedback and dynamic correction process ensures the long-term high fidelity of the digital twin model, enabling it to continuously improve itself by aligning with physical reality, fundamentally guaranteeing the reliability of its predictions. The reliability provides a highly credible data foundation for the entire optimization process, thereby improving the efficiency of the collaborative optimization of geometric parameters for tapered hole machining tools. Furthermore, it innovatively introduces the Pareto stability index in the post-optimization processing stage. A multi-dimensional evaluation system is constructed using this index, coupling the theoretical performance score calculated based on multi-criteria decision-making methods (such as TOPSIS) with the performance fluctuation analysis results based on manufacturing tolerances through a preset function. The performance fluctuation analysis quantifies the sensitivity of candidate solutions to manufacturing errors. This addresses the problem in existing technologies where Pareto optimal solutions often only pursue theoretical performance while neglecting the potential for a sharp performance decline due to parameter fluctuations during actual manufacturing, significantly improving machining quality.

[0163] To solve the problem of huge computational resource consumption in high-fidelity simulation, a phased adaptive learning strategy is designed to guide the training of computational surrogate model. In the global exploration phase, the high-fidelity model is called to simulate in the information sparse area to maximize the reduction of global prediction uncertainty of the model. When the uncertainty falls below the preset threshold, the system automatically switches to the local exploitation phase, in which the expected improvement (EI) is maximized to perform fine sampling in the current superior area. This intelligent active learning strategy ensures that each expensive simulation calculation serves the most urgent information demand, realizes the goal of constructing a high-precision surrogate model with the least number of simulations, and significantly reduces the computational resource consumption and time cost. A mechanism of knowledge management and transfer learning is established. After each optimization task is completed, the knowledge assets of all high-fidelity sample points and the final computational surrogate model are structured and archived to form a reusable knowledge base, which can provide a data basis for further improving the processing quality.

[0164] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for the coordinated optimization of the geometric parameters of a taper hole machining tool, characterized in that, Comprise: S1. Construct a high-fidelity digital twin model with the taper hole machining tool geometry parameters as input, the model is based on the finite element method, integrating the tool-workpiece interaction model of multi-physical field coupling of cutting force-cutting heat-deformation and the dynamic characteristic model of machine tool-clamp-tool-workpiece system, through multi-physical field coupling solving simulation output the prediction value of size accuracy, surface roughness and machining efficiency corresponding to the geometry parameters, generate geometry parameter-performance data mapping set; based on the mapping set as training samples, the calculation agent model is trained; S2. Based on the calculation agent model and the preset multi-element performance objective function, perform multi-objective global optimization, output the candidate Pareto optimal solution set; the preset multi-element performance objective function is based on machining efficiency, surface roughness and size accuracy; S3. Select an initial candidate solution set from the candidate Pareto optimal solution set, and construct a geometry parameter constraint range based on the initial candidate solution set; call the digital twin model based on the geometry parameter constraint range, and construct a preset single objective function for local optimization search to obtain the optimal geometry parameter combination; the single objective function is a comprehensive evaluation function based on surface roughness and size accuracy.

2. The method according to claim 1, wherein: The geometry parameters at least include tool nose radius, opening R angle, rake angle and relief angle; the performance indicators at least include size accuracy, surface roughness and machining efficiency.

3. The method according to claim 2, wherein: The high-fidelity digital twin model also includes a self-correction mechanism, specifically including: Manufacture entity tools based on the geometry parameters, obtain real physical data in cutting experiments, including but not limited to cutting force, temperature and workpiece deformation; Compare the real physical data with the simulation prediction value of the high-fidelity digital twin model, calculate the deviation; Use system identification algorithm to automatically correct the physical parameters in the high-fidelity digital twin model, including but not limited to material constitutive relation and friction coefficient, according to the deviation; Update the corrected model parameters to the geometry parameter-performance data mapping set for training the calculation agent model.

4. The method according to claim 1, wherein: The training process of the calculation agent model is a phased adaptive learning, specifically including: Global exploration stage: based on the initial sample point combination, use the method based on uncertainty sampling for preliminary training, take maximizing reducing the global prediction uncertainty of the model as the criterion, iteratively select new sample points and call the high-fidelity digital twin model for simulation, until the uncertainty decline rate is lower than the first threshold set based on the calculation resource constraint; Local utilization stage: automatically switch to this stage, take the maximum expected improvement as the criterion, iteratively select new sample points in the current excellent area and perform simulation, until the expected improvement is lower than the second threshold set based on the accuracy requirement, form the final calculation agent model; the agent model is based on neural network construction.

5. The method according to claim 4, wherein: The specific acquisition process of the candidate Pareto optimal solution set is: Use the second generation non-dominated sorting genetic algorithm, based on the population size and iteration number of calculation resource optimization, perform global multi-objective optimization on the calculation agent model, generate a preliminary Pareto solution set; The DBSCAN clustering algorithm based on density is applied to the preliminary Pareto solution set to identify and output multiple separate and high-density performance clusters as the input of the initial candidate solution set by adaptively adjusting the density radius and minimum point number.

6. The method of coordinating and optimizing the geometric parameters of a taper hole machining tool according to claim 1, characterized in that: The construction process of the geometric parameter constraint range includes: Statistical analysis of the initial candidate solution set to extract the distribution range of each geometric parameter, including but not limited to mean, variance, and boundary value; Based on the distribution range, combined with manufacturing tolerance and processing equipment constraints, generate the feasible constraint range of geometric parameters; Within the constraint range, call the high-fidelity digital twin model for simulation verification.

7. The method according to claim 6, wherein: The local optimization search adopts a gradient-based optimization method, specifically including: Select the centroid point of the performance cluster from the initial candidate solution set as the initial solution, and call the high-fidelity digital twin model for simulation; Using the adjoint method of the digital twin model, calculate the real physical gradient of performance to geometric parameters through the finite element adjoint solver; Based on the physical gradient, perform local optimization to generate the physical optimal solution as the optimal geometric parameter combination.

8. The method of coordinating and optimizing the geometric parameters of a taper hole machining tool according to claim 1, characterized in that: The method also includes post-processing and knowledge management steps, specifically including: Introduce the Pareto stability index, couple the performance score calculated based on the multi-criteria decision method with the performance fluctuation analysis results within the manufacturing tolerance range through a preset function, and output the geometric parameter combination with the highest PSI index; Structurally archive the knowledge assets formed after the optimization task is completed, including all high-fidelity sample points and the final computational surrogate model, as a reusable knowledge base; When performing a new workpiece material, similar optimization task, load the knowledge assets as a pre-trained model and prior knowledge base to initialize the optimization process of the new task; Compare the experimental data of physical cutting tools with the predicted values of the high-fidelity digital twin model to analyze the model deviation and optimize the performance prediction of subsequent tasks.

Images