Mold structure design optimization method and apparatus

Abstract

The present application relates to the technical field of mold structure design. Disclosed are a mold structure design optimization method and apparatus. The method comprises: performing parametric modeling on a three-dimensional model of a laser cutting die to obtain an adjustable parameter set; performing cutting die geometric feature extraction and feature classification to obtain a feature classification result; performing adaptive multi-scale mesh division to obtain a multi-scale finite element analysis model; performing multi-physics coupling analysis to obtain stress distribution data, deformation data and temperature field distribution data; performing variance analysis to obtain target impact parameters, and, on the basis of the target impact parameters, constructing a multi-objective optimization model; by means of a non-dominated sorting genetic algorithm, solving the multi-objective optimization model to obtain a Pareto optimal solution set; and determining, from the Pareto optimal solution set, target optimization structural parameters of the laser cutting die, thereby improving the optimization efficiency while ensuring the calculation accuracy, and achieving the overall performance improvement of the laser cutting die.

Description

Optimization design method and device of mold structure TECHNICAL FIELD

[0001] The present application relates to the technical field of mold structure design, and particularly relates to an optimization design method and device of mold structure. BACKGROUND

[0002] The structural design of a laser cutter directly affects machining precision and production efficiency. Traditional design methods have been difficult to meet the multi-objective optimization requirements under complex working conditions. In practical applications, laser cutters often face multi-physical field coupling problems such as high temperature, high stress, and large deformation. The interaction between these factors makes the optimization of cutter structure extremely complex.

[0003] At present, the design of laser cutters mainly relies on engineering experience and repeated experiments, and lacks systematic and intelligent optimization methods. This design method not only consumes time and effort, but also makes it difficult to fully consider the trade-off between multiple performance indicators. In addition, traditional finite element analysis methods often face the dilemma of low computational efficiency and insufficient accuracy when dealing with complex geometries and multi-scale problems. SUMMARY

[0004] The present application provides an optimization design method and device of mold structure, which can improve the optimization efficiency while ensuring the calculation accuracy, and realize the overall improvement of the performance of laser cutter.

[0005] The first aspect of the present application provides an optimization design method of mold structure, which comprises:

[0006] Parametric modeling of a three-dimensional model of a laser cutter to obtain a set of adjustable parameters;

[0007] Extracting and classifying the geometric features of the three-dimensional model of the laser cutter to obtain a feature classification result;

[0008] Based on the set of adjustable parameters and the feature classification result, performing adaptive multi-scale grid division on the three-dimensional model of the laser cutter to obtain a multi-scale finite element analysis model;

[0009] Performing multi-physical field coupling analysis on the multi-scale finite element analysis model to obtain stress distribution data, deformation data, and temperature field distribution data;

[0010] Performing variance analysis on the stress distribution data, the deformation data, and the temperature field distribution data to obtain target influence parameters, and constructing a multi-objective optimization model according to the target influence parameters;

[0011] The multi-objective optimization model is solved by using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set, and a target optimization structure parameter of the laser tooling die is determined from the Pareto optimal solution set.

[0012] The second aspect of the application provides a mold structure optimization design device, the mold structure optimization design device comprises:

[0013] A modeling module is configured to perform parameterized modeling on a three-dimensional model of a laser tooling die to obtain a set of adjustable parameters.

[0014] A feature classification module is configured to perform tooling die geometric feature extraction and feature classification on the three-dimensional model of the laser tooling die to obtain a feature classification result.

[0015] A mesh division module is configured to perform adaptive multi-scale mesh division on the three-dimensional model of the laser tooling die based on the set of adjustable parameters and the feature classification result to obtain a multi-scale finite element analysis model.

[0016] A coupling analysis module is configured to perform multi-physical field coupling analysis on the multi-scale finite element analysis model to obtain stress distribution data, deformation data, and temperature field distribution data.

[0017] A construction module is configured to perform variance analysis on the stress distribution data, the deformation data, and the temperature field distribution data to obtain target influence parameters, and to construct a multi-objective optimization model according to the target influence parameters.

[0018] A solution module is configured to solve the multi-objective optimization model by using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set, and to determine a target optimization structure parameter of the laser tooling die from the Pareto optimal solution set.

[0019] The third aspect of the application provides an electronic device, comprising a memory and at least one processor, the memory has instructions stored therein; the at least one processor invokes the instructions in the memory to enable the electronic device to perform the mold structure optimization design method described above.

[0020] The fourth aspect of the application provides a computer-readable storage medium, the computer-readable storage medium has instructions stored therein, when the instructions are run on a computer, the computer performs the mold structure optimization design method described above.

[0021] Compared with the prior art, the application has the following beneficial effects: by parameterizing modeling and geometric feature extraction of the laser cutter die, a set of adjustable parameters and feature classification results are established, which provide a comprehensive and accurate data basis for subsequent optimization, and are beneficial to improve the pertinence and efficiency of optimization. Based on the feature classification results and the set of adjustable parameters, adaptive multi-scale grid division is realized, which can perform fine processing in key areas while taking into account the calculation efficiency, thereby improving the accuracy and efficiency of finite element analysis. By constructing the thermal-mechanical coupling control equation, comprehensive analysis of the laser cutter die under complex working conditions is realized, and stress distribution, deformation and temperature field data can be obtained at the same time, which provides comprehensive performance evaluation for optimization. The variance analysis method is used to identify key influence parameters, and a multi-objective optimization model is constructed, which can effectively reduce the complexity of the optimization problem while considering multiple performance indicators, and improve the optimization efficiency. The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem, a series of Pareto optimal solutions can be obtained, which provides multiple trade-off schemes for decision makers, and enhances the practicality and flexibility of the optimization results. Through comprehensive scoring and decision analysis of the Pareto optimal solution set, the best balance point between multiple optimization objectives can be found, and the final selected optimization structure parameters can meet the performance requirements. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0023] The structures, proportions, sizes, etc. shown in the drawings of the present specification are only used to cooperate with the content disclosed in the specification, to enable those skilled in the art to understand and read, and are not used to limit the conditions that the present application can be implemented, so they do not have technical significance. Any modification of structure, change of proportion relationship or adjustment of size, which does not affect the effects and purposes that the present application can produce, should still fall within the scope of the technical content disclosed by the present application.

[0024] Fig. 1 is a flowchart of the optimization design method of the mold structure provided by the embodiment of the present application;

[0025] Fig. 2 is a structural schematic block diagram of the optimization design device of the mold structure provided by the embodiment of the present application. DETAILED DESCRIPTION

[0026] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only some of the embodiments of the present application, but not all of them. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0027] The flowcharts shown in the drawings are only examples, and do not necessarily include all contents and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be further decomposed, combined or partially merged, so the actual execution order can be changed according to actual situations.

[0028] It should also be understood that the terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application and the appended claims, unless otherwise clear from context, the singular forms "a", "an" and "the" are intended to include the plural forms as well.

[0029] It should be further understood that the term "and / or" used in the present application and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations. Referring to FIG. 1, one embodiment of the mold structure optimization design method in the present application includes:

[0030] Step 100: parameterized modeling of a three-dimensional model of a laser cutter mold to obtain an adjustable parameter set;

[0031] It can be understood that the execution subject of the present application can be a mold structure optimization design device, and can also be a terminal or a server, and the specific embodiments are not limited herein. The present application will be described with the server as an example of the execution subject.

[0032] Specifically, the three-dimensional model of the laser cutter die is geometrically decomposed, and the structure of the laser cutter die is divided into three main parts: the cutter die body, the cutting edge, and the connecting structure. Independent parameterization processing of these parts helps better control and optimize the overall performance of the laser cutter die. When parameterizing the cutter die body, size parameterization processing is performed on the cutter die body. The thickness, length, and width of the cutter die body are key size parameters, and adjustment of these parameters will directly affect the strength and stability of the cutter die. At the same time, the parameterization processing of the cutting edge focuses on its shape characteristics, including the angle of the cutting edge, the radius of the cutting edge, and the length of the cutting edge. The parameterization processing of the connecting structure focuses on the optimization of its layout, and the diameter, position, and number of the connecting holes are the main parameters. By reasonably configuring these parameters, the stability of the cutter die installation and the service life are ensured. According to the material attribute database of the cutter die, the material parameters are extracted and defined. The material properties of the cutter die directly affect its mechanical properties and durability, and the elastic modulus, Poisson's ratio, and density are important material parameters that must be extracted and defined. The machining process specification of the cutter die is analyzed to obtain the surface roughness and heat treatment parameters. According to the use environment specification of the cutter die, the environmental conditions are quantified to obtain the working temperature range and humidity range parameters. The setting of environmental parameters can ensure the reliability and stability of the cutter die in actual use. At the same time, the mechanical performance indicators of the cutter die are quantitatively analyzed to obtain the maximum allowable stress and maximum allowable deformation parameters, which ensure that the cutter die does not fail under high load working conditions. The constraint relationship matrix between each parameter is established to clearly define the dependency and mutual influence between different parameters, and the parameter constraint equation set is obtained. All parameters and parameter constraint equation sets are integrated and normalized. Different dimension parameters are unified into a standardized framework to obtain the final adjustable parameter set.

[0033] Step 200, performing cutter die geometric feature extraction and feature classification on the three-dimensional model of the laser cutter die to obtain a feature classification result;

[0034] Specifically, the three-dimensional model of the laser cutter die is voxelized. By converting the three-dimensional model of the laser cutter die into voxel representation data, a discrete representation of the model is achieved. Voxelization decomposes the three-dimensional model into a series of small cubic units, allowing the geometric structure of the model to be expressed in a format that is more easily processed by a computer. The voxel representation data is processed by a three-dimensional convolutional neural network to extract local features of the model. The three-dimensional convolutional neural network extracts local feature maps of the cutter die geometry through multiple layers of convolution operations. These feature maps can capture local shape features in the model, such as edges, corners, and surface curvature, and other important information. The local feature maps are subjected to max-pooling operations. Max-pooling operations effectively reduce the dimensionality of the feature maps while preserving key feature information by selecting the maximum value in each region of the feature map. The reduced dimensionality feature maps are subjected to non-linear activation function operations to enhance the non-linear expression ability of the features. Activation functions such as ReLU can introduce non-linear factors, allowing the neural network to better learn complex geometric shape features and obtain activated feature maps. Based on the activated feature maps, curvature analysis is performed on the surface of the cutter die. Curvature analysis can identify the trend of changes in the surface of the cutter die and obtain curvature distribution data, which can reflect the degree of bending and shape features of the cutter die surface. Cluster analysis is performed on the curvature distribution data, and similar curvature regions are classified into the same category through clustering operations to form preliminary feature regions. Boundary detection is performed on the preliminary feature regions to identify the precise boundaries of each feature region and obtain feature boundary data. By analyzing the feature boundary data, adjacent feature regions are merged to integrate similar or related regions into larger, more representative feature regions, resulting in optimized feature regions. Shape descriptor calculation is performed on the optimized feature regions. Shape descriptors are a mathematical representation of geometric shape features that can convert complex geometric shape information into feature description vectors. The feature description vectors are classified according to pre-set classification criteria to obtain feature classification results.

[0035] The voxel representation data is input into the input layer of the three-dimensional convolutional neural network for four-dimensional tensor format conversion to meet the requirements of network input and obtain network input data for subsequent convolution operation. The network input data is processed by a first convolutional layer. A three-dimensional convolution operation is performed using 32 3x3x3 convolution kernels, the convolution step is set to 1, the padding is set to 1, the convolution operation is performed on each position of the original data, and the spatial size of the input data is kept unchanged through the padding operation, and a first layer feature map is obtained. In order to ensure that the data distribution in the feature map is more uniform and to accelerate the training of the network, the first layer feature map is subjected to batch normalization processing. Batch normalization can reduce the internal covariate shift and improve the stability of the model. The normalized feature map is subjected to nonlinear activation by applying a ReLU activation function, so that the model can better capture complex geometric features. The activated first layer feature map is subjected to a maximum pooling operation, using a 2x2x2 pooling kernel and a step size of 2. After the pooling operation, the spatial dimension of the feature map is halved, and a first layer down-sampled feature map is obtained. The maximum pooling not only effectively reduces the size of the feature map, but also retains the significance of the local features. The first layer down-sampled feature map is subjected to a second convolutional layer processing. In this stage, a three-dimensional convolution operation is performed using 64 3x3x3 convolution kernels, the step size is also set to 1, and the padding is set to 1, so as to ensure that the feature map after convolution still maintains the same spatial size, and a second layer feature map is obtained. Similar to the first layer processing, the second layer feature map is subjected to batch normalization processing, and a ReLU activation function is applied again to enhance the nonlinear feature expression capability. The activated second layer feature map is also subjected to a maximum pooling operation, using the same 2x2x2 pooling kernel and a step size of 2. After the pooling operation, a second layer down-sampled feature map is obtained. After this layer of processing, the size of the feature map is further reduced, but the significance of the features is enhanced. The second layer down-sampled feature map is subjected to a third convolutional layer processing, a three-dimensional convolution operation is performed using 128 3x3x3 convolution kernels, the step size is still 1, the padding is 1, the size of the feature map after convolution is kept unchanged, and a third layer feature map is obtained. The feature map is subjected to batch normalization processing to ensure the uniformity of the data distribution, and a ReLU activation function is applied to enhance the expression capability of the network. A global average pooling operation is performed on the third layer activated feature map, the spatial dimension of the feature map is compressed to 1x1x1 through global average pooling, and a local feature map is obtained. The local feature map integrates the results of multiple layer processing of the three-dimensional convolutional neural network, and contains key information for deep feature extraction of the input data.

[0036] Step 300, based on the adjustable parameter set and the feature classification result, performing adaptive multi-scale grid division on the three-dimensional model of the laser knife mold to obtain a multi-scale finite element analysis model;

[0037] It should be noted that according to the feature classification result, the three-dimensional model of the laser cutter is regionally divided, and the model is decomposed into multiple important regions with different characteristics. According to the division result, an initial mesh size distribution map is obtained by assigning an initial mesh size to each region. In combination with the adjustable parameter set of the blade parameter, the local mesh refinement is performed on the blade region. Due to the complex geometric characteristics of the blade region and the higher accuracy requirement of the model, finer mesh division is adopted in this region. Through the refined mesh processing of the blade region, more accurate finite element analysis results are obtained in the high stress concentration area. Based on the initial mesh size distribution map and the refined mesh of the blade region, a multi-level mesh division framework is constructed. This framework realizes the multi-scale description of the model by using different mesh densities in different regions. The multi-level mesh division framework is adaptively refined to improve the fineness of the mesh, and a preliminary refined mesh model is obtained. Adaptive refinement is a process of dynamically adjusting the mesh density according to the geometric characteristics and stress distribution of the model, to ensure sufficient mesh density in the key area to capture the details of stress change. The preliminary refined mesh model is subjected to local mesh reconstruction, and the quality and structure of the mesh are optimized to eliminate possible mesh distortion, and an optimized mesh model is obtained. The boundary layer mesh is generated for the optimized mesh model, especially in fluid or heat conduction analysis, the boundary layer region usually has significant gradient change, so it is necessary to generate boundary layer mesh with high resolution to accurately capture these changes. Through this operation, the mesh model with boundary layer is obtained, which ensures that the results of finite element analysis are more accurate in the area with complex boundary conditions. The mesh model with boundary layer is subjected to transition zone mesh generation and node numbering optimization to obtain a multi-scale finite element analysis model. In the transition zone mesh generation process, the fine boundary layer mesh is smoothly transitioned with the relatively coarse internal mesh to avoid numerical errors caused by mesh mutation. Node numbering optimization aims to improve the computational efficiency of finite element analysis. The optimized node numbering can reduce the randomness of memory access in the calculation process and improve the overall performance of the analysis.

[0038] Step 400, multi-physical field coupling analysis is performed on the multi-scale finite element analysis model to obtain stress distribution data, deformation data and temperature field distribution data;

[0039] Specifically, material properties are assigned to the multi-scale finite element analysis model. According to the material characteristics of different regions, each finite element unit is given corresponding material parameters such as elastic modulus, Poisson's ratio, thermal conductivity and specific heat capacity, and a material property distribution model is constructed. According to the working conditions of the laser cutter, the material property distribution model is subjected to corresponding loads and boundary conditions. The loads and boundary conditions include mechanical loads, thermal loads and fixed constraints, which together determine the stress and deformation behavior of the model under actual working conditions. By applying loads and boundary conditions, an initial condition model is constructed. The initial condition model is subjected to thermal-mechanical coupling control equation construction. The heat conduction equation is established to describe the heat transfer process in the material; then the thermal stress equation is established to describe the relationship between stress and strain caused by temperature change, and the coupled equation set is obtained. The coupled equation set is subjected to time discretization processing, by discretizing the continuous time domain into several time steps, the discretized equation set is obtained, which is used to solve the temperature field and stress field changes in each time step. The discretized equation set is subjected to spatial discretization processing, the equation set is projected onto the nodes and elements of the finite element grid, and the global stiffness matrix and load vector are obtained. The global stiffness matrix reflects the stiffness characteristics of each part of the model, and the load vector represents the mechanical response under external load. Based on the global stiffness matrix and load vector, numerical solution is carried out, and the displacement field and temperature field of the model are obtained in each time step. The strain field is calculated, the stress field is calculated through the strain distribution data and the constitutive relation of the material, and the stress distribution data is obtained. The stress distribution data can reveal the stress concentration and possible failure area inside the laser cutter under different loads and temperature conditions. At the same time, the displacement field is post-processed, the node displacement information is extracted, and the deformation data is obtained, which shows the deformation state of the model at each time step. For the temperature field, interpolation processing is carried out to obtain smoother and continuous temperature field distribution data, which reflects the temperature distribution of the model under thermal load.

[0040] Step 500, variance analysis is performed on the stress distribution data, deformation data and temperature field distribution data to obtain target influence parameters, and a multi-objective optimization model is constructed according to the target influence parameters;

[0041] Specifically, data preprocessing is performed on stress distribution data, deformation data, and temperature field distribution data. The stress, deformation, and temperature field distribution data are standardized by converting the data to a unified scale, eliminating the dimensional differences between different physical quantities, and obtaining a standardized data set. Dimensionality reduction is performed on the standardized data set. Through principal component analysis or other dimensionality reduction techniques, high-dimensional data is converted into a low-dimensional feature matrix, retaining the most representative features in the data, reducing the influence of redundant information, and improving the efficiency and accuracy of subsequent analysis. Variance analysis is performed on the feature matrix after dimensionality reduction to calculate the contribution of each feature to the optimization target. Through variance analysis, the importance of each feature in stress, deformation, and temperature and other key performance indicators is quantified, and the feature importance ranking is obtained accordingly. The feature importance ranking reflects the degree of influence of different parameters on the performance of the mold. According to the obtained feature importance ranking, the top K features with the highest contribution are selected to obtain the target influence parameters. Multiple linear regression analysis is performed on the selected target influence parameters to establish a linear relationship model between these parameters and the optimization target, and a preliminary regression equation is obtained. The preliminary regression equation reveals the influence of each target influence parameter on the stress, deformation, and temperature field distribution by fitting the data. Residual analysis is performed on the preliminary regression equation to analyze the differences between the model predicted values and the actual data, correct the bias in the regression model, and obtain a more accurate corrected regression equation. Based on the corrected regression equation, the objective function of the multi-objective optimization problem is constructed, which usually includes minimizing the maximum stress, minimizing the maximum deformation, and minimizing the maximum temperature. Through the integration of the objective function, a multi-objective function expression is obtained. According to the mold design specifications and process limitations, the value range and constraint conditions of each design variable are determined. Through reasonable setting of the design variables, a constraint equation set is obtained, and the multi-objective function expression is combined with the constraint equation set to form a comprehensive objective function. The comprehensive objective function not only considers the needs of multi-objective optimization, but also incorporates actual design and process constraints, making the optimization results more practical and feasible. Sensitivity analysis is performed on the comprehensive objective function to evaluate the influence of each design variable on the optimization result. Sensitivity analysis can identify the most sensitive variables to the optimization result, helping the optimization model to more accurately reflect the actual situation. Finally, a scientific and reasonable multi-objective optimization model is constructed.

[0042] The dimensionality-reduced feature matrix is subjected to data grouping. According to different types of features, the data is divided into stress groups, deformation groups, and temperature groups, forming three independent grouped data sets. Descriptive statistical analysis is performed on the grouped data sets, and statistical characteristics such as mean, variance, and standard deviation of each group of data are calculated to obtain the statistical characteristic description of each group. Based on the statistical characteristic description, normality test is performed on each group of data to determine whether the data conforms to the normal distribution. After obtaining the distribution characteristics of the data, data transformation is performed to adjust the data to an appropriate range or distribution form, obtaining the transformed data set. This may include logarithmic transformation, square root transformation, or other forms of nonlinear transformation on the data to make the data closer to the normal distribution. The transformed data set is subjected to homogeneity of variance test to determine whether the variances of different data groups are equal, obtaining the inter-group variance result. According to the result of the homogeneity of variance test, if a group of data satisfies the homogeneity of variance, single-factor analysis of variance (ANOVA) is used to process the data; while for those data groups with heterogeneous variances, Welch's ANOVA is used for analysis. The result of ANOVA gives the F statistic and the corresponding p value of each group of data, and the F statistic measures the significance of the inter-group difference, while the p value provides the statistical significance judgment standard. Multiple comparisons are made on the F statistic and the p value to determine the significance level of the difference between different groups. Based on the significance level, the effect size of each feature is calculated, reflecting the explanatory power of each feature on the target variable. By processing the effect size, the explained variance proportion of each feature is obtained, indicating the size of the contribution of each feature to the overall variance. The explained variance proportion is subjected to normalization processing to obtain the relative contribution of each feature to the target. According to the relative contribution, all features are ranked in descending order to obtain the feature importance ranking.

[0043] Step 600, the multi-objective optimization model is solved by non-dominated sorting genetic algorithm, and a Pareto optimal solution set is obtained. From the Pareto optimal solution set, the target optimization structure parameters of the laser cutter die are determined.

[0044] Specifically, the multi-objective optimization model is encoded. The design variables are converted into binary strings for processing in the genetic algorithm. Through encoding, an initial population is generated, and these initial individuals serve as the starting point for the optimization process. The initial population is evaluated for fitness, and a fitness matrix is formed by calculating the fitness value of each individual under each objective function. Based on the fitness matrix, the population is sorted non-dominantly. Non-dominant sorting layers the individuals according to their performance in the multi-objective function, resulting in the non-dominant level of each individual. The lower the non-dominant level, the better the individual performs in the optimization objective. On this basis, the individuals of the same non-dominant level are calculated for crowding degree, which calculates the density of each individual in the solution space to obtain the crowding degree value. Individuals with larger crowding degree values indicate better diversity in the solution space and should therefore be prioritized in the selection operation. Based on the non-dominant level and crowding degree value, the selection operation selects individuals with high fitness and low crowding degree to form the parent population. After selection, the parent population undergoes crossover operation, which combines the genetic information of the parent individuals to generate new offspring individuals, resulting in a post-crossover offspring population. Crossover operation explores better solution space regions by recombining existing solutions. Polynomial mutation operation is performed on the post-crossover offspring population to introduce random perturbations, increase population diversity, and avoid local optimal solutions. The mutated offspring population is combined with the parent population to form a larger merged population. The merged population contains all the information of the parent and offspring populations, ensuring that excellent individuals are not lost due to generational replacement. Selection operation is performed on the merged population to gradually eliminate underperforming individuals until the population size returns to the predetermined size. The selection process continues until the maximum number of iterations or convergence conditions are met. Convergence conditions usually occur when the algorithm fails to find significantly better solutions within several generations, indicating that the algorithm has reached an optimal state. After meeting these conditions, the Pareto optimal solution set is obtained, containing a set of non-dominated solutions, each representing the optimal trade-off between different objectives. Decision analysis is performed on the Pareto optimal solution set by calculating the comprehensive score of each solution to evaluate its performance in different objectives. The calculation of the comprehensive score can be based on actual needs, weighting the importance of each objective to obtain the comprehensive performance score of each solution. The solution with the highest comprehensive score is selected as the target optimization structure parameter of the laser cutter.

[0045] In the embodiments of the present application, by parameterizing modeling and geometric feature extraction of the laser cutter die, a set of adjustable parameters and feature classification results are established, providing a comprehensive and accurate data basis for subsequent optimization, which is conducive to improving the pertinence and efficiency of optimization. Based on the feature classification results and the set of adjustable parameters, adaptive multi-scale meshing is realized, which can perform fine processing in key areas while taking into account the computational efficiency, improving the accuracy and efficiency of finite element analysis. By constructing the thermal-mechanical coupling control equation, comprehensive analysis of the laser cutter die under complex working conditions is realized, and stress distribution, deformation and temperature field data can be obtained simultaneously, providing a comprehensive performance evaluation for optimization. The variance analysis method is used to identify key influencing parameters, and a multi-objective optimization model is constructed, which can effectively reduce the complexity of the optimization problem while considering multiple performance indicators, improving the optimization efficiency. The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem, which can obtain a series of Pareto optimal solutions, providing multiple trade-off schemes for decision makers, enhancing the practicality and flexibility of the optimization results. Through comprehensive scoring and decision analysis of the Pareto optimal solution set, the best balance point between multiple optimization objectives can be found, ensuring that the final selected optimization structure parameters meet the performance requirements.

[0046] In a specific embodiment, the process of step 100 can specifically include the following steps:

[0047] The geometric structure of the three-dimensional model of the laser cutter die is decomposed to obtain the cutter die body, the cutting edge and the connecting structure;

[0048] The cutter die body is subjected to size parameterization processing to obtain the thickness, length and width parameters, the cutting edge is subjected to shape parameterization processing to obtain the cutting edge angle, cutting edge radius and cutting edge length parameters, and the connecting structure is subjected to layout parameterization processing to obtain the connecting hole diameter, position and number parameters;

[0049] According to the cutter die material attribute database, the material parameters are extracted and defined to obtain the elastic modulus, Poisson's ratio and density parameters, and the cutter die processing technology specification is analyzed to obtain the surface roughness and heat treatment parameters;

[0050] According to the cutter die use environment specification, the environmental conditions are quantified to obtain the working temperature range and humidity range parameters, and the mechanical performance indicators of the cutter die are quantitatively analyzed to obtain the maximum allowable stress and maximum allowable deformation parameters;

[0051] A constraint relationship matrix between each parameter is established to obtain a parameter constraint equation set, and all parameters and the parameter constraint equation set are integrated and normalized to obtain a set of adjustable parameters.

[0052] Specifically, the overall geometry of the laser knife die is decomposed. The complex three-dimensional model is decomposed into multiple functionally explicit parts, including the knife die body, the cutting edge, and the connecting structure. The knife die body is subjected to size parameterization processing. The knife die body is the core part of the entire structure, and its geometric size directly affects the overall performance and working effect of the knife die. In the parameterization process, the main parameters include the thickness, length, and width of the body. For example, if the thickness of the knife die body is denoted as , the length as , and the width as , these parameters can be established based on specific design requirements or standard specifications. For example, for a certain material or working condition, the body thickness may be required to vary within a certain range to adapt to different manufacturing needs. The thickness of the body becomes an adjustable parameter, and its range can be represented as , where and are the minimum and maximum allowed values of the thickness, respectively. At the same time, the shape of the cutting edge is crucial to the cutting performance of the knife die. When performing shape parameterization processing on the cutting edge, key parameters such as the cutting edge angle, cutting edge radius, and cutting edge length are considered. Assuming that the cutting edge angle is denoted as , the cutting edge radius as , and the cutting edge length as , the selection of these parameters needs to be determined based on the type and thickness of the cutting material. For example, a larger cutting edge angle may be suitable for materials with higher hardness, while a smaller cutting edge angle is suitable for softer materials. The selection of the cutting edge radius and the cutting edge length should also be optimized according to the requirements of cutting accuracy and knife die life. For the connecting structure part, its main function is to ensure the stable connection of the knife die with other equipment or components. The layout parameterization processing of the connecting structure needs to focus on parameters such as the diameter, position, and number of connecting holes. Assuming that the diameter of the connecting hole is denoted as , the position as , and the number as , the determination of these parameters needs to consider the strength requirements of mechanical connection and the convenience of installation. The range of the connecting hole diameter usually needs to be defined according to the specifications of the bolts or pins used, while the position The distribution needs to be reasonable according to the geometry of the tool die and the use scenario to ensure the best connection effect. According to the tool die material attribute database, the material parameters are extracted and defined. The material properties of the tool die are important factors affecting its mechanical properties, usually including basic material parameters such as elastic modulus, Poisson's ratio and density. Assuming that the elastic modulus is denoted as , the Poisson's ratio is denoted as , and the density is denoted as , the selection of these parameters needs to be matched with the working environment and load conditions of the tool die. For example, for a high-strength and high-rigidity tool die, the elastic modulus needs to be higher, and the Poisson's ratio should be within a suitable range to ensure that the tool die is not prone to fracture or permanent deformation under high stress conditions. Analyze the processing technology specifications of the tool die. The processing technology directly affects the surface quality and service life of the tool die. Important processing technology parameters include surface roughness and heat treatment parameters, which can improve the wear resistance and fatigue resistance of the tool die. For high-precision cutting requirements, the surface roughness should be as small as possible to reduce friction loss and improve cutting accuracy. According to the use environment specifications of the tool die, the environmental conditions are quantified. This includes determining the working temperature range and humidity range of the tool die in actual use. Assuming that the working temperature range is denoted as , and the humidity range is denoted as , these environmental parameters need to be considered in combination with material parameters to ensure that the tool die can maintain stable performance under harsh environmental conditions. For example, the material of the tool die may need to have a low thermal expansion coefficient at high temperatures to prevent changes in dimensional accuracy. Quantitatively analyze the mechanical performance indicators of the tool die. Important mechanical performance indicators include maximum allowable stress and maximum allowable deformation. Assuming that the maximum allowable stress is denoted as , and the maximum allowable deformation is denoted as , the determination of these indicators needs to consider the working conditions of the tool die under extreme load conditions. The maximum allowable stress is usually limited by the yield strength or breaking strength of the material, while the maximum allowable deformation needs to be determined through experiments or simulations to ensure that the tool die does not undergo excessive deformation or failure during use. Establish the constraint relationship matrix between the parameters and construct the parameter constraint equation set. Assuming that the form of a certain parameter constraint equation is:

[0053] ;

[0054] where represents the constraint condition, is each parameter, is the weight coefficient, is a constant term. The constraint relationship matrix and the equation set help to ensure that all parameters can work together while meeting the design specifications and avoid conflicts between parameters. In order to improve the optimization efficiency, all parameters and parameter constraint equations are integrated and normalized. Through normalization, parameters of different dimensions can be unified to the same scale to obtain an adjustable parameter set.

[0055] In a specific embodiment, the process of performing step 200 can specifically include the following steps:

[0056] The three-dimensional model of the laser tool is voxelized to obtain voxel representation data;

[0057] The voxel representation data is processed by a three-dimensional convolutional neural network to obtain a local feature map;

[0058] The local feature map is subjected to a max-pooling operation to obtain a reduced dimension feature map, and the reduced dimension feature map is subjected to a nonlinear activation function operation to obtain an activated feature map;

[0059] According to the activated feature map, curvature analysis is performed on the surface of the tool to obtain curvature distribution data, and clustering analysis is performed on the curvature distribution data to obtain a preliminary feature region;

[0060] Boundary detection is performed on the preliminary feature region to obtain feature boundary data, and according to the feature boundary data, adjacent feature regions are merged to obtain an optimized feature region;

[0061] The optimized feature region is subjected to shape descriptor calculation to obtain a feature description vector, and the feature description vector is classified to obtain a feature classification result.

[0062] Specifically, the three-dimensional model is converted into voxel representation data. Voxelization is a process of discretizing continuous three-dimensional geometric data, which divides three-dimensional space into multiple small cubic units, i.e. voxels, each of which represents the properties of a small volume in the model. Assuming that the three-dimensional model of the laser tool contains voxels, these voxels can be represented by a three-dimensional matrix, where each matrix element corresponds to the attribute value of a voxel. Voxelization converts complex three-dimensional geometric shapes into discrete data structures that are easy to calculate and analyze, facilitating subsequent deep learning processing. The voxel representation data is processed by a three-dimensional convolutional neural network. The three-dimensional convolutional neural network extracts local feature maps of the model through multiple convolutional layers. Convolution operation can be regarded as a filter sliding in three-dimensional space, which extracts the features of the local region through point product operation. Assuming that a convolutional layer uses filters with a size of The output feature map can be represented as:

[0063] ;

[0064] wherein, is the weight parameter of the th convolution kernel, is the value of the voxel representation data, is the bias term, is the output feature map after convolution operation. Through multi-layer convolution and step-by-step stacking of feature maps, the three-dimensional convolutional neural network can capture the multi-level local features of the model and extract important information about the geometry of the tooling. Max-pooling operation is performed on the local feature map to reduce the spatial dimension of the feature map. Max-pooling is a dimension reduction technique that selects the maximum value in the pooling window to represent the features of the region, achieving data compression. Assuming the size of the pooling window is 2, the feature map after pooling can be represented as:

[0065] ;

[0066] Max-pooling operation reduces the size of the feature map and retains the most significant feature information, improving the computational efficiency of the model. Nonlinear activation function operation is performed on the dimension-reduced feature map, and commonly used activation functions such as RelU can enhance the nonlinear expression ability of the model. Based on the activated feature map, curvature analysis of the tooling surface is performed. Curvature is an important geometric property that measures the degree of surface bending and can reflect subtle shape changes on the tooling surface. Assuming the curvature distribution in space is The curvature distribution data is obtained by calculating the principal curvatures of each point in the feature map. The curvature analysis can help identify high-curvature regions on the die surface, which usually correspond to parts with sharp cutting edges or shape mutations. Cluster analysis is performed on the curvature distribution data to group similar curvature regions into the same class. Through the clustering algorithm, preliminary feature regions are identified, which usually represent parts of the die with similar geometric characteristics, such as blade tips, corners, etc. The result of the cluster analysis is to divide the die surface into several feature regions, each corresponding to a certain class of geometric features. Boundary detection is performed on the preliminary feature regions. By identifying the transition regions between regions, the boundaries of each feature region are determined, and feature boundary data is obtained. By analyzing the boundary data, adjacent feature regions are merged to integrate similar or related regions into larger, more representative optimized feature regions. For example, in the blade region of the die, boundary detection can help identify the edges and edges of the blade, and these edge regions are merged into a whole feature region. Shape descriptor calculation is performed on the optimized feature regions. Shape descriptors are mathematical tools for quantitatively describing geometric features, which can extract complex geometric shape information into a vector. Suppose a shape descriptor vector is represented as , where each represents a shape attribute of the feature region, such as area, average curvature, or boundary length, etc. The shape descriptor vector is classified, and each feature region is classified into different geometric feature categories. The classification process can be completed by machine learning algorithms such as support vector machines or k-nearest neighbor classifiers. Through classification, feature classification results are obtained. For example, in the blade region, the shape descriptor may show that the region has high curvature and sharp edges, which can be used to judge the sharpness of the blade and the cutting performance.

[0067] In a specific embodiment, the process of performing step of performing three-dimensional convolutional neural network processing on the voxel representation data to obtain the local feature map can specifically include the following steps:

[0068] The voxel representation data is input into the input layer of the three-dimensional convolutional neural network for four-dimensional tensor format conversion to obtain network input data;

[0069] The network input data is processed by a first convolutional layer using 32 3x3x3 convolutional kernels for three-dimensional convolution operation with a step size of 1 and a padding of 1 to obtain a first layer feature map;

[0070] The first layer feature map is processed by batch normalization and an ReLU activation function to obtain a first layer activation feature map, and the first layer activation feature map is subjected to a maximum pooling operation using a 2x2x2 pooling kernel with a step size of 2 to obtain a first layer down-sampling feature map;

[0071] The first-layer downsampled feature map is processed by the second convolutional layer. 64 3x3x3 convolutional kernels are used to perform three-dimensional convolution operations with a stride of 1 and padding of 1 to obtain the second-layer feature map.

[0072] Batch normalization is performed on the second-layer feature map, and the ReLU activation function is applied to obtain the second-layer activated feature map. Max pooling is then performed on the second-layer activated feature map using a 2x2x2 pooling kernel with a stride of 2 to obtain the second-layer downsampled feature map.

[0073] The second-layer downsampled feature map is processed by the third convolutional layer. 128 3x3x3 convolutional kernels are used to perform three-dimensional convolution operation with a stride of 1 and padding of 1 to obtain the third-layer feature map.

[0074] Batch normalization is performed on the third-layer feature map, and the ReLU activation function is applied to obtain the third-layer activation feature map. Global average pooling is then performed on the third-layer activation feature map to compress its spatial dimension to 1x1x1, resulting in a local feature map.

[0075] Specifically, the voxel representation data is input into the input layer of the 3D convolutional neural network and then transformed into a four-dimensional tensor format. The original 3D voxel data is expanded into a four-dimensional tensor format with multiple channels to match the input requirements of the convolutional neural network. Assume the voxel data is of size... A three-dimensional matrix, in which , and These represent depth, height, and width, respectively. To accommodate the input layer of the neural network, this 3D data is converted... The four-dimensional tensor format, in which It refers to the batch size (i.e., the number of samples processed at one time). This is the number of channels (which can be set to 1 for single-voxel data), while , and Keep it unchanged. Process the transformed network input data using the first convolutional layer. Use 32... The convolution kernel performs 3D convolution operations. The core of convolution operations lies in extracting features from local regions by performing weighted summation on the input data through a sliding window.

[0076] Assuming the input data is ,in Indicates the first One convolutional kernel, , and These represent the depth, height, and width of the convolution kernel, respectively, and the output feature map after convolution. may be expressed as:

[0077] ;

[0078] wherein, is the bias term of the th convolution kernel, , and is the index of the output feature map after convolution. The step length of the convolution kernel is set to 1 and the padding is set to 1. The convolution operation calculates every position of the input data, while maintaining the spatial size of the output feature map through the padding operation. The first layer feature map is processed by batch normalization. By normalizing the values of the feature map, it has the same mean and variance, eliminating the statistical distribution difference of different batches of data, improving the stability and convergence speed of network training. The feature map after batch normalization is then applied to the ReLU activation function for nonlinear activation. The expression of the ReLU function is:

[0079] ;

[0080] This operation truncates all negative values to 0, thereby introducing a nonlinear factor that enables the network to better fit complex data patterns. The activated feature map retains important features in the input data while suppressing irrelevant information. The first layer of activated feature map is processed by max-pooling operation. Max-pooling selects the maximum value within the pooling window to represent the features of the region, thereby realizing dimension reduction and data compression of the feature map. Assuming the size of the pooling window is and the step length is 2, the feature map after pooling can be expressed as:

[0081] ;

[0082] The max-pooling operation reduces the size of the feature map by half, obtaining the first layer down-sampling feature map. The first layer down-sampling feature map is processed by the second convolution layer. A three-dimensional convolution operation is performed using 64 convolution kernels, with steps similar to the first convolution layer. The step length of the convolution kernel is still set to 1 and the padding is set to 1. After the second convolution operation, the second layer feature map is obtained. The second layer feature map is processed by batch normalization and applied to the ReLU activation function to ensure that the model maintains the ability to express nonlinear features. After activation, the feature map is processed by max-pooling operation, using the same pooling kernel and step length as the first layer, reducing the spatial size of the feature map to obtain the second layer down-sampling feature map. The second layer down-sampling feature map is processed by the third convolution layer. In this layer, 128 The third layer of convolution is performed by a three-dimensional convolution operation with a larger number of convolution kernels. The third layer of convolution can capture more complex geometric features and higher-level abstract representations through a larger number of convolution kernels and deeper network layers. The output of the third layer of feature map contains key information in the three-dimensional structure of the laser cutter die. After batch normalization processing and ReLU activation of the third layer of feature map, the third layer of activation feature map is obtained. The global average pooling operation is performed on the third layer of activation feature map. The global average pooling compresses the spatial dimension of the feature map to , that is, the average value of all values of each feature map is calculated and taken as the output of the feature map. Assuming that the size of the third layer of activation feature map is , then the output of the global average pooling is expressed as:

[0083]

[0084] Through the global average pooling, the network compresses the spatial information into representative local feature maps, providing highly integrated feature expression for the final output of the model.

[0085] In a specific embodiment, the process of performing step 300 can specifically include the following steps:

[0086] According to the feature classification result, the three-dimensional model of the laser cutter die is regionally divided and the initial grid size is allocated, and an initial grid size distribution map is obtained;

[0087] According to the blade parameter in the adjustable parameter set, the local grid refinement is performed on the blade region to obtain a blade region refined grid, and a multi-level grid division framework is constructed based on the initial grid size distribution map and the blade region refined grid;

[0088] The multi-level grid division framework is adaptively refined to obtain a preliminary refined grid model, and the preliminary refined grid model is locally reconstructed to obtain an optimized grid model;

[0089] The boundary layer grid of the optimized grid model is generated to obtain a grid model with a boundary layer, and the grid model with a boundary layer is subjected to transition zone grid generation and node number optimization to obtain a multi-scale finite element analysis model.

[0090] Specifically, the three-dimensional model of the laser toolpath is regionally partitioned according to the feature classification results. The three-dimensional model of the toolpath is partitioned into multiple regions, each of which can have different physical properties and geometric complexities. For example, the toolpath body, the cutting edge, and the connecting region are partitioned into different regions, and the mesh partitioning strategy for each region can be different to meet specific computational needs. After the regional partitioning is completed, the initial mesh size is assigned to each region to form an initial mesh size distribution map. The initial mesh size assignment is usually based on the geometric properties of the region and the key parameters contained in the feature classification results. For example, regions with higher geometric complexity can require smaller mesh sizes to capture subtle geometric details. According to the cutting edge parameters in the adjustable parameter set, the cutting edge region is locally refined. As the most critical part of the toolpath, the accuracy of the cutting edge directly affects the cutting performance of the toolpath, so more detailed mesh partitioning is required in this region. For example, if the geometric parameters of the cutting edge include the cutting edge angle , the cutting edge radius , and the cutting edge length , the degree of mesh refinement is determined according to these parameters. The angle of the cutting edge Smaller time can mean that the region has a larger curvature, requiring a denser mesh to capture the subtle changes. The process of mesh refinement is achieved by dividing the initial mesh into smaller elements, which can more accurately describe the shape and boundaries of the cutting edge. Refining the mesh in the cutting edge region can significantly improve the analysis accuracy without increasing the overall computational load. Combining the refined mesh in the cutting edge region with the initial mesh size distribution map, a multi-level mesh partitioning framework is constructed. The multi-level mesh partitioning framework achieves a multi-scale description of the entire model by using different mesh densities in different regions. Simpler geometric regions can retain larger mesh elements, while geometrically complex or stress-concentrated regions use finer meshes, reducing the overall computational cost while maintaining high accuracy. On the basis of the multi-level mesh partitioning framework, adaptive refinement is performed to obtain a preliminary refined mesh model. Adaptive refinement is a process of automatically adjusting the mesh density based on the preliminary calculation results. By identifying regions with large stress, deformation, or temperature gradients in the model, the mesh in these regions is refined. For example, in the preliminary calculation of finite element analysis, stress concentration in a certain region may cause local error to increase. Adaptive refinement can dynamically adjust the mesh density in this region to improve the accuracy of the calculation. The preliminary refined mesh model is subjected to local mesh reconstruction to eliminate possible mesh distortion or low-quality elements, resulting in an optimized mesh model. The process of mesh reconstruction adjusts the shape and distribution of mesh elements to improve the quality indicators of the elements, such as the shape factor and area ratio. The optimized mesh model better adapts to complex geometries while avoiding error accumulation in numerical analysis. For example, by adjusting the position of certain mesh nodes, the shape of adjacent elements is made more regular, improving the overall mesh quality and stability of the analysis. The optimized mesh model is subjected to boundary layer mesh generation. Boundary layer mesh is suitable for fluid dynamics or heat conduction problems, where the gradient of physical quantities such as velocity or temperature is large near the solid surface. Boundary layer mesh generates a denser mesh layer near the surface to accurately capture these gradient changes. For example, in the thermal analysis of a laser tool, the temperature gradient near the cutting edge surface is large, so a dense boundary layer mesh is needed in this area to ensure the accuracy of the heat conduction analysis. The mesh model with boundary layers is subjected to transition zone mesh generation and node numbering optimization to obtain the final multi-scale finite element analysis model. The purpose of transition zone mesh generation is to smoothly transition between the dense boundary layer mesh and the coarse main body mesh, thereby avoiding numerical errors caused by sudden changes in mesh size. Node numbering optimization improves the computational efficiency of finite element analysis by optimizing the numbering order of mesh nodes. Assuming that the stiffness matrix and the load vector of the finite element equation system are:

[0091] ;

[0092] wherein, is the nodal displacement vector, the nodal numbering optimization reduces the distribution of non-zero elements in matrix and improves the speed and efficiency of matrix solution. The final multi-scale finite element analysis model can meet the requirements of accuracy and efficiency at the same time.

[0093] In a specific embodiment, the process of performing step 400 can specifically include the following steps:

[0094] Material property distribution model is obtained by assigning corresponding elastic modulus, Poisson's ratio, thermal conductivity and specific heat capacity parameters to each element according to the material properties of different regions.

[0095] According to the working conditions of the laser cutter, load and boundary conditions are applied to the material property distribution model, including mechanical load, thermal load and fixed constraint, to obtain an initial condition model.

[0096] The initial condition model is subjected to thermal-mechanical coupling control equation construction, thermal conduction equation and thermal stress equation are established, coupled equation groups are obtained, and time discretization processing is performed on the coupled equation groups to obtain discretized equation groups.

[0097] The discretized equation groups are subjected to spatial discretization processing to obtain global stiffness matrix and load vector, and displacement field and temperature field at each time step are solved according to the global stiffness matrix and load vector.

[0098] Strain calculation is performed on the displacement field to obtain strain distribution data, and stress field is calculated according to the strain distribution data and material constitutive relation to obtain stress distribution data.

[0099] Post-processing is performed on the displacement field to extract nodal displacement information and obtain deformation data, and interpolation processing is performed on the temperature field to obtain temperature field distribution data.

[0100] Specifically, each element is assigned corresponding physical parameters based on the material properties of different regions, including elastic modulus, Poisson's ratio, thermal conductivity, and specific heat capacity, etc. The material properties of different regions directly affect the mechanical and thermal behavior of the model. By assigning appropriate material parameters to each finite element in the model, a material property distribution model reflecting the real material behavior is constructed. According to the working conditions of the laser cutter, load and boundary conditions are applied to the material property distribution model. Loading conditions usually include mechanical load, thermal load, and fixed constraints, etc. For example, in actual use, the laser cutter may bear cutting force as mechanical load, and at the same time, it may also be affected by thermal load due to the heat generated during cutting. Fixed constraints are applied to some key positions to simulate the fixed support of the cutter in the actual working state. By applying these conditions, the initial condition model is obtained. The construction of the thermal-mechanical coupling control equation is carried out on the initial condition model. The core of thermal-mechanical coupling analysis is to consider the mutual influence of heat conduction and thermal stress. The heat conduction equation describes the diffusion behavior of temperature in the material, and its mathematical expression is:

[0101] ;

[0102] where, is the density, is the specific heat capacity, is the temperature, is the thermal conductivity, is the internal heat source term. The equation shows that temperature change is affected not only by the thermal properties of the material itself, but also by external heat sources and heat conduction effects. At the same time, the thermal stress equation describes the stress distribution caused by temperature change, and its basic form is:

[0103] ;

[0104] where, is the stress tensor, is the body force density. The thermal stress equation combines the thermal expansion effect and the mechanical response of the material, reflecting the influence of thermal load on the internal stress state of the material. By coupling the heat conduction equation and the thermal stress equation, a complete set of coupled equations is obtained, which together describe the behavior of the material under the action of thermal-mechanical coupling. In order to solve these coupled equations on the computer, time discretization is performed. The continuous time domain is divided into discrete time steps, so that the temperature and stress state of each time step can be solved step by step. Common methods include forward Euler method or implicit backward Euler method, and the time step size The choice of the time step size is crucial for the stability and accuracy of the solution. Through time discretization, a set of discrete equations is obtained, and the temperature and stress at each time step can be solved step by step. On the basis of time discretization, spatial discretization is performed on the discrete equations. Spatial discretization is to expand the coupled equations on the entire finite element mesh to obtain the global stiffness matrix and the load vector

[0105] ;

[0106] where, is the displacement vector, represents the stiffness matrix of the material, which integrates the elastic modulus, Poisson's ratio, and geometric information; and contains the external load applied to the model. By solving the linear equations, the displacement field and temperature field at each time step are obtained. Strain calculation is performed on the displacement field. Strain calculation is based on the gradient of the displacement field. Assuming that the strain tensor is , the relationship between it and the displacement gradient is:

[0107] ;

[0108] The strain tensor describes the geometric properties of material deformation, combined with the material constitutive relationship (such as Hooke's law for linear elastic materials), the stress field

[0109] ;

[0110] where, is the elastic matrix of the material, reflecting parameters such as the elastic modulus and Poisson's ratio of the material. Through this process, the stress distribution data inside the model is obtained, revealing the stress state of the material under the action of the load. Post-processing is performed on the displacement field to extract the displacement information of the nodes, obtaining the deformation data of the entire model. Deformation data shows the displacement of each part of the laser tool under the combined action of external load and thermal effect. For example, if the deformation of the tool exceeds the allowable range of design, it may lead to failure in actual use, thereby affecting the processing quality. Interpolation is performed on the temperature field to obtain more accurate temperature field distribution data. Interpolation can obtain continuous temperature distribution by interpolating data between discrete calculation nodes. For example, through the temperature field distribution data, the thermal fatigue of the laser tool during work is evaluated, and the design is optimized accordingly to prolong its service life.

[0111] In a specific embodiment, the process of performing step 500 can specifically include the following steps:

[0112] Data preprocessing is performed on the stress distribution data, deformation data, and temperature field distribution data to obtain a standardized dataset, and dimensionality reduction is performed on the standardized dataset to obtain a feature matrix after dimensionality reduction;

[0113] An analysis of variance is performed on the feature matrix after dimensionality reduction to calculate the contribution of each feature to the target, and a feature importance ranking is obtained;

[0114] According to the feature importance ranking, the top K features with the highest contribution are selected to obtain target influence parameters, and a multiple linear regression analysis is performed on the target influence parameters to establish a linear relationship model between the factors and the target, and a preliminary regression equation is obtained;

[0115] Residual analysis is performed on the preliminary regression equation to obtain a corrected regression equation, and based on the corrected regression equation, a target function of a multi-objective optimization problem is constructed, including minimization of maximum stress, minimization of maximum deformation, and minimization of maximum temperature, to obtain a multi-objective function expression;

[0116] According to the mold design specifications and process restrictions, the value range and constraint conditions of each design variable are determined to obtain a constraint equation system, and the multi-objective function expression and the constraint equation system are combined to obtain a comprehensive objective function;

[0117] Sensitivity analysis is performed on the comprehensive objective function to obtain a multi-objective optimization model.

[0118] Specifically, the stress distribution data, deformation data, and temperature field distribution data are standardized, and the mean normalization or z-score standardization method is used to subtract the mean of each data feature and divide by the standard deviation, eliminating the dimensional differences between different features to obtain a standardized dataset. The standardized dataset is subjected to dimensionality reduction processing to reduce the dimensionality of the data and extract the most representative features. Common dimensionality reduction techniques include principal component analysis and linear discriminant analysis. Through dimensionality reduction processing, high-dimensional data is converted into a low-dimensional feature matrix, thereby retaining most of the effective information of the data and eliminating redundancy. An analysis of variance is performed on the feature matrix after dimensionality reduction to calculate the contribution of each feature to the optimization target. The purpose of variance analysis is to quantify the role of each feature in explaining the changes in the target variable, and to rank these features according to their contribution. By analyzing the variance contribution of each feature, it is identified which features play a key role in the optimization process. Assuming that the variance contribution of a certain feature is , it can be calculated by the following formula:

[0119] ;

[0120] wherein represents the variance of feature . is the total number of features. By calculating the variance contribution of each feature and sorting them, the feature importance ranking is obtained. According to the feature importance ranking, the top features with the highest contribution are selected, resulting in the target impact parameters. Multiple linear regression analysis is performed on the target impact parameters to establish a linear relationship model between the factors and the optimization targets. Multiple linear regression analysis describes the relationship between the target variable and the selected features by fitting a linear model. Assuming the target variable is , and the feature vector is (containing selected features), the preliminary regression equation can be expressed as:

[0121]

[0122] where is the intercept term, is the regression coefficient, is the error term. The value of is estimated by the least squares method, obtaining the preliminary regression equation, which reveals the linear relationship between the target variable and each feature. Residual analysis is performed on the preliminary regression equation to evaluate the fitting effect and accuracy of the model. Residual analysis includes calculating the mean and variance of the residuals, and detecting the normality and independence of the residuals. If the residual analysis shows systematic bias or non-normal distribution, the regression model needs to be corrected. By adjusting the parameters in the regression model or introducing nonlinear terms, the corrected regression equation is obtained. Based on the corrected regression equation, the objective function of the multi-objective optimization problem is constructed. For the optimization design of laser knife mold, the objective function usually includes multiple objectives such as minimizing the maximum stress, minimizing the maximum deformation, and minimizing the highest temperature. Assuming these objectives are represented as , and , the expression of the multi-objective function can be expressed as:

[0123]

[0124] where represents the vector of design variables, containing all design parameters related to the optimization targets. The multi-objective function expression reflects the need to optimize multiple objectives within the design space. To ensure that the optimization results meet the actual design specifications and process limitations, determine the value range and constraint conditions of each design variable, obtaining the constraint equation set. For example, a design variable may need to satisfy the following constraints:

[0125]

[0126] where and​​​ the minimum and maximum values of the variable The multi-objective function expression is combined with these constraints to obtain a comprehensive objective function. The comprehensive objective function not only considers the optimization objectives, but also incorporates actual design and process constraints, ensuring that the optimization results are practically operable. Sensitivity analysis is performed on the comprehensive objective function to assess the degree of influence of different design variables on the optimization results. The results of the sensitivity analysis can help identify which variables are most critical to the final design scheme and how to adjust these variables to optimize the design effect. Finally, an accurate multi-objective optimization model is constructed to guide the design optimization process of the laser cutter die.

[0127] In a specific embodiment, the process of performing variance analysis on the reduced dimension feature matrix to calculate the contribution of each feature to the target and obtaining the feature importance ranking can specifically include the following steps:

[0128] The reduced dimension feature matrix is grouped into data, and the data is divided into stress group, deformation group and temperature group according to the feature type, to obtain a grouped data set;

[0129] Descriptive statistical analysis is performed on the grouped data set to obtain statistical feature description, and normality test is performed on each group of data based on the statistical feature description to obtain data distribution characteristics;

[0130] Based on the data distribution characteristics, data transformation is performed to obtain a transformed data set, and homogeneity of variance test is performed on the transformed data set to obtain inter-group variance results;

[0131] According to the inter-group variance results, single factor ANOVA is used for variance homogeneity groups, and Welch's ANOVA is used for variance inhomogeneity groups to obtain F statistics and p values of each group;

[0132] Multiple comparisons are performed on the F statistics and p values to obtain the significance level of the inter-group difference, and based on the significance level, the effect size of each feature is calculated to obtain the explained variance proportion of each feature;

[0133] The explained variance proportion is normalized to obtain the relative contribution of each feature to the target, and the features are arranged in descending order according to the relative contribution to obtain the feature importance ranking.

[0134] Specifically, the reduced dimension feature matrix is grouped by feature type. According to the physical meaning represented by the feature, it is divided into three groups: stress group, deformation group and temperature group, to obtain the grouped data set. Descriptive statistical analysis is performed on the grouped data set to obtain the basic statistical characteristics of the data. Descriptive statistical analysis includes calculating the mean, standard deviation, median, skewness and kurtosis of each group of data, which helps to identify the central tendency, dispersion and distribution form of the data. According to the statistical characteristic description, normality test is performed on each group of data to determine whether the data conforms to the normal distribution. Commonly used normality test methods include Shapiro-Wilk test or Kolmogorov-Smirnov test. The results of normality test will provide the basis for the selection of subsequent statistical analysis methods. If the data deviates from the normal distribution, further data transformation may be needed. Based on the data distribution characteristics, the data is transformed accordingly to obtain the transformed data set that meets the analysis requirements. Common data transformation methods include logarithmic transformation, square root transformation or Box-Cox transformation, etc. These methods can help the data to be closer to the normal distribution. For example, if the stress group data presents a right-skewed distribution, it can be transformed by logarithm to pull it closer to the normal distribution:

[0135] ;

[0136] wherein, is the transformed data. After data transformation, a new transformed data set is obtained. Variance homogeneity test is performed on the transformed data set. Variance homogeneity test is used to determine whether the variances of different groups of data are equal. Commonly used variance homogeneity test methods include Levene test and Bartlett test. If the test result shows that the variances are homogeneous, that is, the variances of each group of data are equal, then single factor analysis of variance (ANOVA) can be used to analyze the differences between groups of data; if the variances are not homogeneous, Welch’s ANOVA should be used, which is more robust to the case of unequal variances. Through single factor ANOVA or Welch’s ANOVA, the F statistic and p value of each group of data are calculated. F statistic reflects the ratio of between-group variation to within-group variation, and p value is used to judge the significance of the statistical result. For example, F statistic can be calculated by the following formula:

[0137] ;

[0138] If the p-value is lower than a preset significance level (usually 0.05), it can be considered that there is a significant difference between different groups. Based on the F-statistic and the p-value, a multiple comparison analysis is performed to explore which features have significant differences between different data groups. Based on the multiple comparison analysis, the effect size of each feature is calculated to obtain the explained variance proportion of each feature. The effect size is used to measure the contribution of the feature to the change in the target variable, and is a key indicator for feature importance ranking. For example, the effect size of a certain feature can be represented as:

[0139]

[0140] where the between-group sum of squares reflects the variation of the feature between different groups, and the total sum of squares reflects the variation of the feature in the overall data. The larger the effect size, the stronger the explanatory power of the feature on the target variable. The explained variance proportion is normalized to obtain the relative contribution of each feature to the target. Normalization converts the effect sizes of all features to values between 0 and 1 in proportion, ensuring that they are compared on the same scale. The relative contribution can be represented as:

[0141]

[0142] where n is the total number of features. According to the relative contribution, all features are ranked in descending order to obtain the importance ranking of the features, thereby identifying the features that have the greatest impact on the target variable. In a specific embodiment, the process of performing step 600 can specifically include the following steps:

[0143] Encode the multi-objective optimization model, convert the design variables into binary strings, obtain the initial population, and evaluate the fitness of the initial population to calculate the fitness value of each individual under each objective function, and obtain the fitness matrix;

[0144] Sort the population based on the fitness matrix to obtain the non-dominated level of each individual, and calculate the crowding degree of individuals with the same non-dominated level to obtain the crowding degree value;

[0145] Select based on the non-dominated level and the crowding degree value to obtain the parent population, and perform crossover operation on the parent population to generate offspring individuals, and obtain the offspring population after crossover;

[0146] Perform polynomial mutation operation on the offspring population after crossover to introduce random disturbance, and obtain the mutated offspring population;

[0147]

[0148] ​​​​The mutated offspring population is combined with the parent population to form a combined population, and selection is performed on the combined population until a maximum number of iterations or a convergence condition is reached, to obtain a Pareto optimal solution set;

[0149] The Pareto optimal solution set is subjected to decision analysis, and a comprehensive score of each solution is calculated, and the solution with the highest comprehensive score is selected as the target optimized structure parameter of the laser cutter die.

[0150] Specifically, the design variables are encoded and converted into binary strings to represent the solutions of the optimization problem. The encoding method helps to discretize the continuous design variables, so as to adapt to the solving process of the genetic algorithm. For example, assuming that there are three design variables , and , each variable can take a value within a specific range. By quantizing the continuous variables into binary form, for example, using an 8-bit binary string to represent each variable, each design solution is represented as a 24-bit binary string. This string is an individual in the genetic algorithm, and the initial population is composed of multiple such binary individuals. The initial population is subjected to fitness evaluation, and the performance of each individual under each objective function is calculated to determine its merits. Assuming that there are two objective functions and , where is the design variable vector obtained by decoding the binary string. Fitness evaluation is performed by decoding the binary individual into continuous variable values, and then substituting them into the objective functions to calculate the fitness values of each individual under these objective functions. For example, if the binary string of an individual is decoded to obtain a value of 3.75, a value of 2.5, and after substituting into the objective functions, it may obtain and . The fitness values of all individuals form a fitness matrix. According to the fitness matrix, the individuals in the population are subjected to non-dominated sorting. Non-dominated sorting is a key step in handling multi-objective optimization in genetic algorithms, and individuals are divided into different levels according to their dominance relationship. If one individual dominates another, it means that it is not worse than the other individual in all objective functions, and at least performs better in one objective function. Through non-dominated sorting, individuals are divided into multiple levels, and the first level individuals are non-dominated, i.e. no individual in the population dominates them. The non-dominated level of each individual is represented by an integer, and the smaller the integer, the better the individual. For individuals of the same non-dominated level, the crowding distance is calculated, and the adjacent distance of each individual in the objective space is measured. The crowding distance reflects the distribution density of the individual in the objective space. For example, if a certain individual is in the objective functions and The values of the upper and lower are 10 and 5, respectively, and the two nearest neighbors of The values of the upper and lower are 8 and 12, respectively, and the two nearest neighbors of The values of the upper and lower are 4 and 6, respectively, and the congestion of an individual can be determined by calculating the sum of these two intervals. The smaller the congestion, the more densely packed the individual is in the target space, and the genetic algorithm will tend to preserve individuals with larger congestion during the selection process to maintain the diversity of the population. Based on the non-dominated rank and congestion values, the parent population is selected. During the selection process, the genetic algorithm first selects individuals with lower non-dominated ranks (i.e., individuals with smaller rank values), and if there are multiple individuals in the same rank, the individual with larger congestion is preferred. This selection strategy ensures that the next generation not only preserves better solutions but also maintains the diversity of solutions. After selection, the parent population is subjected to crossover operations to generate offspring individuals. Crossover operations generate new offspring individuals by exchanging parts of the binary string of two parent individuals. For example, two binary strings 10101011 and 01010100 can generate two new individuals 10100100 and 01011011 through crossover. The offspring population after crossover is subjected to polynomial mutation operations. Mutation is an important step in genetic algorithms that introduces random perturbations to allow the population to escape local optimal solutions and more comprehensively explore the solution space. Polynomial mutation is typically implemented by randomly flipping certain bits in the binary string. For example, individual 10101011 can become 10111011 after mutation, and small-scale random mutation can enhance the global search ability of the algorithm. The mutated offspring population is combined with the parent population to form a larger merged population. By selecting the merged population, the less performing individuals are gradually eliminated, and the individuals with high fitness and strong diversity are preserved until the size of the population returns to the initial population size. This process is repeated until the maximum number of iterations is reached or the convergence condition is met. The convergence condition usually includes the change in the objective function being less than a certain threshold or the fitness of the optimal solution in the population no longer improving significantly. The solutions obtained through this process constitute the Pareto optimal solution set, which represents the best trade-off between different objectives in multi-objective optimization. Decision analysis is performed on the Pareto optimal solution set to select the final optimal solution. In decision analysis, weights are assigned to each objective function, and the overall score of each solution is calculated based on the weights. For example, assuming the weights of the objective functions and are and , the overall score of a solution can be represented as:

[0151] ;

[0152] By calculating a comprehensive score for all solutions, the solution with the highest score is selected as the final optimization solution. In the design of a laser cutter die, this optimization process can help determine the geometric parameters of the die, material selection, and process conditions, so that the die achieves the best balance in multiple performance indicators such as stress, deformation, and temperature control. For example, through this optimization process, the optimal size and material combination of the die under different working conditions are found under design constraints, thereby extending the service life of the die and improving cutting accuracy.

[0153] The above describes the optimization design method of the mold structure in the embodiments of the present application, and the following describes the optimization design device 10 of the mold structure in the embodiments of the present application. Referring to FIG. 2, one embodiment of the optimization design device 10 of the mold structure in the embodiments of the present application includes:

[0154] The modeling module 11 is configured to perform parameterized modeling on the three-dimensional model of the laser cutter die to obtain a set of adjustable parameters.

[0155] The feature classification module 12 is configured to perform cutter die geometric feature extraction and feature classification on the three-dimensional model of the laser cutter die to obtain a feature classification result.

[0156] The mesh division module 13 is configured to perform adaptive multi-scale mesh division on the three-dimensional model of the laser cutter die based on the set of adjustable parameters and the feature classification result to obtain a multi-scale finite element analysis model.

[0157] The coupling analysis module 14 is configured to perform multi-physical field coupling analysis on the multi-scale finite element analysis model to obtain stress distribution data, deformation data, and temperature field distribution data.

[0158] The construction module 15 is configured to perform variance analysis on the stress distribution data, deformation data, and temperature field distribution data to obtain target influence parameters, and construct a multi-objective optimization model according to the target influence parameters.

[0159] The solving module 16 is configured to perform non-dominated sorting genetic algorithm solving on the multi-objective optimization model to obtain a set of Pareto optimal solutions, and determine the target optimization structure parameters of the laser cutter die from the set of Pareto optimal solutions.

[0160] Through the cooperation of the above-mentioned components, through the parameterized modeling and geometric feature extraction of the laser cutter die, the adjustable parameter set and feature classification result are established, which provides a comprehensive and accurate data basis for subsequent optimization, and is beneficial to improve the pertinence and efficiency of optimization. Based on the feature classification result and the adjustable parameter set, adaptive multi-scale mesh division is realized, which can perform fine processing in key areas while considering the calculation efficiency, improving the accuracy and efficiency of finite element analysis. By constructing the thermal-mechanical coupling control equation, the comprehensive analysis of the laser cutter die under complex working conditions is realized, and the stress distribution, deformation and temperature field data can be obtained at the same time, which provides a comprehensive performance evaluation for optimization. The variance analysis method is used to identify the key influence parameters, and a multi-objective optimization model is constructed, which can effectively reduce the complexity of the optimization problem while considering multiple performance indicators, and improve the optimization efficiency. The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem, which can obtain a series of Pareto optimal solutions, provide multiple trade-off schemes for decision makers, and enhance the practicality and flexibility of the optimization results. Through comprehensive scoring and decision analysis of the Pareto optimal solution set, the best balance point between multiple optimization objectives can be found to ensure that the finally selected optimization structure parameters meet the performance requirements.

[0161] The application also provides an electronic device, which comprises a memory and a processor, the memory stores computer readable instructions, and the computer readable instructions are executed by the processor to make the processor execute the steps of the optimization design method of the mold structure in each of the above embodiments.

[0162] The application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium, and the computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the optimization design method of the mold structure.

[0163] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0164] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0165] The above description and the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of optimal design of a mold structure, characterized by, The optimization design method of the mold structure comprises: Performing parameterized modeling on a three-dimensional model of the laser cutter mold to obtain an adjustable parameter set; Performing cutter mold geometric feature extraction and feature classification on the three-dimensional model of the laser cutter mold to obtain a feature classification result; Performing adaptive multi-scale grid division on the three-dimensional model of the laser cutter mold based on the adjustable parameter set and the feature classification result to obtain a multi-scale finite element analysis model; Performing multi-physics field coupling analysis on the multi-scale finite element analysis model to obtain stress distribution data, deformation data and temperature field distribution data; Performing variance analysis on the stress distribution data, the deformation data and the temperature field distribution data to obtain target influence parameters, and constructing a multi-objective optimization model according to the target influence parameters; Solving the multi-objective optimization model by using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set, and determining target optimization structure parameters of the laser cutter mold from the Pareto optimal solution set.

2. The method of optimizing a mold structure according to claim 1, wherein The parameterized modeling on the three-dimensional model of the laser cutter mold to obtain an adjustable parameter set comprises: Performing geometric structure decomposition on the three-dimensional model of the laser cutter mold to obtain a cutter body, a cutting edge and a connecting structure; Performing size parameterization processing on the cutter body to obtain body thickness, length and width parameters, shape parameterization processing on the cutting edge to obtain cutting edge angle, cutting edge radius and cutting edge length parameters, and layout parameterization processing on the connecting structure to obtain connecting hole diameter, position and number parameters; Extracting and defining material parameters according to a cutter mold material attribute database to obtain elastic modulus, Poisson's ratio and density parameters, and analyzing cutter mold processing technology specifications to obtain surface roughness and heat treatment parameters; Quantifying environmental conditions according to cutter mold usage environment specifications to obtain working temperature range and humidity range parameters, and quantitatively analyzing cutter mold mechanical performance indicators to obtain maximum allowable stress and maximum allowable deformation amount parameters; Establishing a constraint relationship matrix among various parameters to obtain a parameter constraint equation set, and integrating and normalizing all parameters and the parameter constraint equation set to obtain an adjustable parameter set.

3. The method of claim 1, wherein The cutter mold geometric feature extraction and feature classification on the three-dimensional model of the laser cutter mold to obtain a feature classification result comprises: Performing voxelization processing on the three-dimensional model of the laser cutter mold to obtain voxel representation data; Performing three-dimensional convolutional neural network processing on the voxel representation data to obtain a local feature map; Performing a max-pooling operation on the local feature map to obtain a reduced dimension feature map, and performing a nonlinear activation function operation on the reduced dimension feature map to obtain an activated feature map; Performing curvature analysis on the cutter mold surface according to the activated feature map to obtain curvature distribution data, and performing clustering analysis on the curvature distribution data to obtain preliminary feature regions; Performing boundary detection on the preliminary feature regions to obtain feature boundary data, and performing merging processing on adjacent feature regions according to the feature boundary data to obtain optimized feature regions; The shape descriptor calculation is performed on the optimized feature region to obtain a feature description vector, and the feature description vector is classified to obtain a feature classification result.

4. The method of claim 3, wherein The three-dimensional convolutional neural network processing of the voxel representation data comprises: The voxel representation data is input into an input layer of a three-dimensional convolutional neural network for four-dimensional tensor format conversion to obtain network input data; The network input data is processed by a first convolutional layer, a three-dimensional convolution operation is performed using 32 3x3x3 convolution kernels, the step length is 1, and the padding is 1 to obtain a first layer feature map; The first layer feature map is processed by batch normalization and a ReLU activation function to obtain a first layer activation feature map, and a maximum pooling operation is performed on the first layer activation feature map using a 2x2x2 pooling kernel and a step length of 2 to obtain a first layer down-sampling feature map; The first layer down-sampling feature map is processed by a second convolutional layer, a three-dimensional convolution operation is performed using 64 3x3x3 convolution kernels, the step length is 1, and the padding is 1 to obtain a second layer feature map; The second layer feature map is processed by batch normalization and a ReLU activation function to obtain a second layer activation feature map, and a maximum pooling operation is performed on the second layer activation feature map using a 2x2x2 pooling kernel and a step length of 2 to obtain a second layer down-sampling feature map; The second layer down-sampling feature map is processed by a third convolutional layer, a three-dimensional convolution operation is performed using 128 3x3x3 convolution kernels, the step length is 1, and the padding is 1 to obtain a third layer feature map; The third layer feature map is processed by batch normalization and a ReLU activation function to obtain a third layer activation feature map, and a global average pooling operation is performed on the third layer activation feature map to compress the spatial dimension to 1x1x1 to obtain a local feature map.

5. The method of claim 1, wherein The adaptive multi-scale grid division of the three-dimensional model of the laser cutter die based on the adjustable parameter set and the feature classification result comprises: According to the feature classification result, the three-dimensional model of the laser cutter die is regionally divided and initially grid size is allocated to obtain an initial grid size distribution map; According to the blade parameter in the adjustable parameter set, a local grid refinement is performed on the blade region to obtain a blade region refined grid, and based on the initial grid size distribution map and the blade region refined grid, a multi-level grid division framework is constructed; The multi-level grid division framework is adaptively refined to obtain a preliminary refined grid model, and the preliminary refined grid model is locally reconstructed to obtain an optimized grid model; The boundary layer grid generation is performed on the optimized grid model to obtain a grid model with a boundary layer, and the grid generation and node number optimization are performed on the grid model with the boundary layer to obtain a multi-scale finite element analysis model.

6. The method of claim 1, wherein The multi-physical field coupling analysis of the multi-scale finite element analysis model comprises: Material property distribution model is obtained by assigning corresponding elastic modulus, Poisson's ratio, thermal conductivity and specific heat capacity parameters to each element according to material characteristics of different regions through material property distribution of the multi-scale finite element analysis model; Initial condition model is obtained by applying load and boundary conditions including mechanical load, thermal load and fixed constraint to the material property distribution model according to working conditions of the laser cutter die; Coupling equation set is obtained by constructing heat conduction equation and thermal stress equation through heat-force coupling control equation construction of the initial condition model, and time discretization processing is performed on the coupling equation set to obtain discretization equation set; Global stiffness matrix and load vector are obtained by performing spatial discretization processing on the discretization equation set, and displacement field and temperature field of each time step are obtained by solving the global stiffness matrix and the load vector; Strain distribution data is obtained by strain calculation of the displacement field, and stress field is calculated according to the strain distribution data and material constitutive relation to obtain stress distribution data; Deformation data is obtained by extracting node displacement information through post-processing of the displacement field, and temperature field distribution data is obtained by interpolation processing of the temperature field.

7. The method of claim 1, wherein The stress distribution data, the deformation data and the temperature field distribution data are subjected to variance analysis to obtain target influence parameters, and a multi-objective optimization model is constructed according to the target influence parameters, including: Standardized data set is obtained by data preprocessing of the stress distribution data, the deformation data and the temperature field distribution data, and dimension reduction processing is performed on the standardized data set to obtain a feature matrix after dimension reduction; Feature importance ranking is obtained by calculating the contribution of each feature to the target through variance analysis of the feature matrix after dimension reduction; The top K features with the highest contribution are selected according to the feature importance ranking to obtain target influence parameters, and a linear relationship model between factors and targets is established through multiple linear regression analysis of the target influence parameters to obtain a preliminary regression equation; The preliminary regression equation is subjected to residual analysis to obtain a corrected regression equation, and a target function of a multi-objective optimization problem is constructed based on the corrected regression equation, including minimization of maximum stress, minimization of maximum deformation and minimization of maximum temperature, to obtain a multi-objective function expression; The value range and constraint conditions of each design variable are determined according to the mold design specification and process limitation to obtain a constraint equation set, and a comprehensive target function is obtained by combining the multi-objective function expression and the constraint equation set; The comprehensive target function is subjected to sensitivity analysis to obtain a multi-objective optimization model.

8. The method of optimization design of a mold structure according to claim 7, wherein The feature matrix after dimension reduction is subjected to variance analysis to calculate the contribution of each feature to the target and obtain feature importance ranking, including: The feature matrix after dimension reduction is subjected to data grouping, and the data is divided into stress group, deformation group and temperature group according to feature type to obtain a grouped data set; Statistical feature description is obtained through descriptive statistical analysis of the grouped data set, and data distribution characteristics are obtained by normality test of each group of data according to the statistical feature description; Perform data transformation based on the data distribution characteristics to obtain a converted data set, and perform variance homogeneity test on the converted data set to obtain inter-group variance results; According to the inter-group variance results, use single-factor ANOVA for groups with variance homogeneity and use Welch's ANOVA for groups with variance inhomogeneity to obtain F statistics and p values of each group; Perform multiple comparisons on the F statistics and p values to obtain the significance level of the inter-group difference, and based on the significance level, calculate the effect size of each feature to obtain the explained variance proportion of each feature; Perform normalization processing on the explained variance proportion to obtain the relative contribution of each feature to the target, and arrange the features in descending order according to the relative contribution to obtain the feature importance ranking.

9. The method of claim 1, wherein The non-dominated sorting genetic algorithm is used to solve the multi-objective optimization model to obtain a Pareto optimal solution set, and the target optimization structure parameters of the laser cutter are determined from the Pareto optimal solution set, including: Encode the multi-objective optimization model to convert design variables into binary strings to obtain an initial population, and evaluate the fitness of the initial population to calculate the fitness value of each individual under each objective function to obtain a fitness matrix; According to the fitness matrix, perform population non-dominated sorting to obtain the non-dominated level of each individual, and calculate the crowding degree of individuals with the same non-dominated level to obtain a crowding degree value; Based on the non-dominated level and the crowding degree value, select to obtain a parent population, and perform crossover operation on the parent population to generate offspring individuals to obtain a cross offspring population; Perform polynomial mutation operation on the cross offspring population to introduce random disturbance to obtain a mutated offspring population; Merge the mutated offspring population and the parent population to form a merged population, and select the merged population until the maximum number of iterations or the convergence condition is reached to obtain a Pareto optimal solution set; Perform decision analysis on the Pareto optimal solution set to calculate the comprehensive score of each solution, and select the solution with the highest comprehensive score as the target optimization structure parameters of the laser cutter.

10. A device for optimizing the design of mold structures, characterized in that, The mold structure optimization design method for performing the mold structure optimization design method according to any one of claims 1-9, the mold structure optimization design device comprises: A modeling module for parameterizing modeling of a three-dimensional model of a laser cutter to obtain a set of adjustable parameters; A feature classification module for extracting and classifying geometric features of the three-dimensional model of the laser cutter to obtain a feature classification result; A mesh division module for performing adaptive multi-scale mesh division on the three-dimensional model of the laser cutter based on the set of adjustable parameters and the feature classification result to obtain a multi-scale finite element analysis model; A coupling analysis module for performing multi-physical field coupling analysis on the multi-scale finite element analysis model to obtain stress distribution data, deformation data, and temperature field distribution data; A coupling analysis module for performing multi-physical field coupling analysis on the multi-scale finite element analysis model to obtain stress distribution data, deformation data, and temperature field distribution data; The construction module is configured to perform variance analysis on the stress distribution data, the deformation data and the temperature field distribution data to obtain target influence parameters, and construct a multi-target optimization model according to the target influence parameters. The solving module is configured to perform non-dominated sorting genetic algorithm solving on the multi-target optimization model to obtain a Pareto optimal solution set, and determine target optimization structure parameters of the laser cutter from the Pareto optimal solution set.

Images