Machine learning-based constitutive parameter identification for crystal plasticity models

Abstract

The crystal plasticity finite element (CPFE) method has emerged as an effective tool in probing the deformation mechanism of crystalline materials. A critical challenge in practical CPFE applications is the rapid and precise calibration of parameters for the crystal plasticity constitutive model, essential for accurate simulations. In this study, the machine learning method is employed to identify the parameters of the crystal plasticity constitutive model. The proposed machine learning method can directly determine the material parameters of the adopted constitutive model from experimentally obtained macroscopic tensile stress-strain curves. A proper Voronoi polycrystalline finite element model is established, and the uniaxial tensile stress-strain curves of polycrystalline copper (Cu) are calculated by using the adopted crystal plasticity constitutive model with different parameter combinations, thereby constructing a database for machine learning. The findings demonstrate that the stress-strain curves simulated from model parameters, as predicted by the machine learning method, align closely with the experimental results. Furthermore, feature importance analysis, utilizing the random forest algorithm, elucidates the relationship between the constitutive model parameters and the macroscopic stress-strain curve characteristics, such as yield stress and strain hardening rate. Additionally, the machine learning model, trained with the simulation data of Cu, is capable of determining the material parameters of other face-centered cubic metals, such as Ni, AISI 316L stainless steel and CrMnFeCoNi HEA, showcasing its extensive utility.