CN-121999939-A - Machine learning identification method for porous material constitutive parameters

CN121999939ACN 121999939 ACN121999939 ACN 121999939ACN-121999939-A

Abstract

The invention provides a machine learning identification method for porous material constitutive parameters, which comprises the steps of constructing a two-dimensional porous finite element model representing sintering nano silver microcosmic porous characteristics, obtaining mechanical response data corresponding to different crystal plasticity constitutive parameters, constructing a data set for constitutive parameter identification, obtaining a trained machine learning model, embedding the trained machine learning model into a differential evolution algorithm respectively, generating a test individual, inputting the test individual into a machine learning agent model, obtaining corresponding prediction mechanical response data, evaluating the matching degree, obtaining an evaluation value, outputting an optimal crystal plasticity constitutive parameter combination, and evaluating the fitting effect of a parameter identification result. According to the invention, under the framework of the crystal plasticity constitutive framework, the efficient identification of the constitutive parameters of the porous material can be realized, the parameter optimization efficiency is remarkably improved, and the calculation resource consumption is reduced.

Inventors

LIU LU
Fu Huacheng
Yu Huachen
LEI MINGQI
Huang Shoukun
CAI ZHIKUANG
WANG ZIXUAN

Assignees

南京邮电大学
南京邮电大学南通研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (9)

1. A machine learning identification method for porous material constitutive parameters is characterized by comprising the following steps, S1, constructing a two-dimensional porous finite element model for representing microscopic porous characteristics of sintered nano silver; S2, importing a two-dimensional porous finite element model into finite element simulation software ABAQUS, and carrying out numerical simulation on mechanical responses under different grain orientation conditions based on a crystal plasticity theory to obtain mechanical response data corresponding to different crystal plasticity constitutive parameters, so as to construct a data set for recognizing the constitutive parameters; S3, respectively training a plurality of machine learning models based on the data set, performing parameter tuning, and establishing a mapping relation between crystal plasticity constitutive parameters and mechanical response data to obtain a trained machine learning model; s4, respectively embedding the trained machine learning model into a differential evolution algorithm, and randomly initializing a population in a preset constitutive parameter range, wherein each individual in the population represents a group of five-dimensional crystal plasticity constitutive parameter vectors; s5, sequentially executing mutation, crossover and selection operations on population individuals based on a differential evolution algorithm to generate test individuals, inputting the test individuals into a machine learning agent model to obtain corresponding predicted mechanical response data, and calculating errors between the predicted mechanical response data and the test stress data based on an evaluation formula to evaluate the matching degree and obtain an evaluation value; s6, repeating the step S5, and continuously carrying out iterative updating on the population according to the evaluation value until the preset iterative condition is met, and outputting the optimal crystal plasticity constitutive parameter combination; and S7, combining and substituting the optimal crystal plasticity constitutive parameters into a two-dimensional porous finite element model for verification, obtaining corresponding verification mechanical response data, and comparing the verification mechanical response data with test mechanical response data to evaluate the fitting effect of the parameter identification result.
2. The method for machine learning and identifying parameters of porous materials according to claim 1, wherein step S1 is specifically, S11, observing a two-dimensional microstructure of the sintered nano silver material by using a scanning electron microscope to obtain a scanning electron microscope image, and extracting structural feature parameters for two-dimensional numerical modeling from the scanning electron microscope image; S12, according to grain order and size distribution characteristics in structural characteristic parameters, utilizing Neper software to generate grain seed points with controlled random distribution in a two-dimensional representative voxel region, and adopting a Voronoi mosaic method to construct a two-dimensional polycrystalline grain geometric structure based on the grain seed points; S13, on the basis of a two-dimensional grain geometric model, on the premise of keeping the consistency of porosity and pore scale statistical characteristics, carrying out equivalent simplification treatment on irregular pore morphology presented in the scanning electron microscope image according to pore morphology parameters and distribution characteristics extracted from the scanning electron microscope image, equivalently representing the irregular pore morphology as regular geometric pores, introducing a pore area, distinguishing a material matrix phase from a pore phase, and constructing a two-dimensional porous finite element model for representing the sintering nano silver micro porous characteristics.
3. The machine learning identification method for the constitutive parameters of the porous material according to claim 2, wherein in step S11, structural characteristic parameters for two-dimensional numerical modeling are extracted from a scanning electron microscope image, specifically, image segmentation and binarization processing are performed on the scanning electron microscope image to identify a material matrix region, a grain boundary and a pore region, the number of grains contained in a two-dimensional representative voxel model is obtained through statistics based on the grain boundary, meanwhile, equivalent circle diameter calculation is performed on the pore region to obtain an average value of pore diameters, and the porosity of sintered nano silver is determined according to the area occupation ratio of the pore region in the image.
4. A machine learning identification method for porous material constitutive parameters according to claim 1-3, wherein in step S2, the crystal plasticity constitutive parameters comprise initial critical splitting stress GIN, young' S modulus EMOD, strain rate sensitivity parameter AN, asymptotic hardening rate THEA1 and initial hardening rate THEAZ.
5. A machine learning identification method for porous material constitutive parameters according to any one of claims 1-3, wherein in step S2, the mechanical response data comprises a final stress and a yield stress, wherein the final stress is a stress-strain curve composed of a plurality of points obtained by simulation of a nano silver constitutive model by finite element simulation software ABAQUS, the final stress is selected according to the stress-strain curve, and the yield stress is a stress value corresponding to 20% of the stress-strain curve.
6. The method for machine learning and identifying parameters of porous material according to any one of claims 1-3, wherein in step S3, a support vector regression model SVR, a long-short-term memory artificial neural network model LSTM and a multi-layer perceptron MLP are adopted as a machine learning model.
7. The method for machine learning identification of constitutive parameters of a porous material according to any one of claims 1 to 3, wherein in step S3, when a plurality of machine learning models are respectively trained by using a data set and parameter tuning is performed, a mean square error MSE and a decision coefficient R2 are adopted as performance evaluation indexes.
8. A machine learning identification method for porous material constitutive parameters according to claim 1-3, wherein in step S5, test stress data is obtained by performing a uniaxial tensile test on a sintered nano silver material, setting a test temperature to 60 ℃, setting a strain rate to 0.001%. S-1, and constructing a strain-stress curve from a plurality of discrete data points, wherein the test stress data comprises a final stress obtained by the test and a yield stress obtained by the test.
9. The method for machine learning identification of porous material constitutive parameters according to any one of claims 1-3, wherein in step S5, an error between predicted mechanical response data and test stress data is calculated based on an evaluation formula to evaluate a degree of matching and obtain an evaluation value D obj : , Wherein, the Is the final stress predicted by the trained machine learning model, Is the final stress obtained by the test, Is the yield stress predicted by the trained machine learning model, The evaluation value describes the maximum difference between the yield stress and the final stress and the test value, and when the evaluation value is minimum, the calculation result is closest to the test value.

Description

Machine learning identification method for porous material constitutive parameters Technical Field The invention relates to a machine learning identification method for constitutive parameters of a porous material, and belongs to the technical field of metal materials. Background The crystal plastic constitutive model is an important theoretical tool for describing plastic deformation behavior of a metal material, can reveal stress-strain response of the material under the action of external load and a microscopic deformation mechanism of the material from grain size, and is widely applied to material processing technology optimization, service performance prediction and finite element numerical simulation. Compared with a compact metal material, the porous metal material has obvious pore structure characteristics, and the grain size, the pore morphology and the spatial distribution of the porous metal material have decisive influence on the mechanical property and the deformation mechanism of the material, so that the constitutive behavior of the porous metal material presents stronger nonlinearity and dimensional dependence, and higher requirements are put forward on crystal plastic modeling and parameter identification. In a crystal plastic constitutive model, accurate identification of model parameters is a key for realizing accurate prediction of material mechanical behaviors. For porous materials, due to their highly heterogeneous microstructure, finite element numerical simulations are often required on a representative voxel basis to characterize the effect of pores on crystal slip, hardening behavior, and overall mechanical response. In the existing research, intelligent optimization methods such as differential evolution algorithm and the like have been introduced into the crystal plasticity parameter inversion process so as to improve the global property and stability of parameter search. However, the method needs to frequently call high-cost finite element calculation in the iterative process, particularly in a porous crystal plastic model, the single simulation calculation amount is large, the convergence is poor, the parameter identification efficiency is low, and the calculation resource consumption is huge. In recent years, machine learning models are increasingly being used to replace traditional finite element calculations due to their advantages in non-linear mapping and high dimensional data fitting to build fast predictive models of mechanical response. By establishing a sample data set in the early stage and training, the method can rapidly predict stress response without complex simulation again. However, most of the prior researches only pay attention to improvement of the machine learning model in terms of prediction precision, and lack of organic combination with a global optimization algorithm, so that acceleration effect of machine learning in parameter optimization cannot be fully exerted. Therefore, aiming at the porous crystal plastic material, the existing parameter identification method still has the problems of low calculation efficiency, insufficient adaptability to complex microstructures, limited intelligent degree and the like, and is difficult to meet the requirement of rapid and accurate intrinsic parameter identification under a complex porous material system. A novel crystal plastic parameter identification method combining a machine learning agent model and a differential evolution algorithm is needed, so that the parameter optimization efficiency is greatly improved while the mechanical response prediction precision of a porous material is ensured, and effective support is provided for crystal plastic modeling and engineering application of the porous material. Disclosure of Invention The invention aims to provide a machine learning identification method for constitutive parameters of porous materials, which solves the problems of low optimization efficiency and high calculation resource consumption in the prior art. The technical scheme of the invention is as follows: A machine learning identification method for porous material constitutive parameters comprises the following steps, S1, constructing a two-dimensional porous finite element model for representing microscopic porous characteristics of sintered nano silver; S2, importing a two-dimensional porous finite element model into finite element simulation software ABAQUS, and carrying out numerical simulation on mechanical responses under different grain orientation conditions based on a crystal plasticity theory to obtain mechanical response data corresponding to different crystal plasticity constitutive parameters, so as to construct a data set for recognizing the constitutive parameters; S3, respectively training a plurality of machine learning models based on the data set, performing parameter tuning, and establishing a mapping relation between crystal plasticity constitutive parameters and mechanical resp