Study On Prediction Method Of Crystal Plastic Constitutive Parameters Based On BP Neural Network Algorithm

Co MPared with macroscopic phenomenological constitutive models,crystal plasticity theory can associate microstructure changes of metal materials with macroscopic plastic deformation,and can effectively explain the internal mechanism of metal material deformation.Therefore,it is widely used in various disciplines.The premise of using crystal plastic finite element method to solve problems is to determine all constitutive parameters therein.The traditional method of determining parameters is to continuously fit macro or micro experimental data and simulation data,and ultimately find the appropriate combination of material parameters.The essence of this method is trial and error method.The solution process is relatively complex,the calculation time is long,the calculation amount is large,the uncertainty is strong,and it is not easy to obtain a parameter combination with good fitting effect.This article combines the BP(Back Prop Association)neural network algorithm to provide a convenient and applicable method for determining the parameters of the constitutive model.Based on the rate dependent phenomenological crystal plasticity theory,this paper established a single crystal plastic finite element model suitable for face centered and body centered cubic crystal structures,and introduced the modeling process in detail.A polycrystalline uniaxial tensile simulation model for two types of crystal structures was established using representative volume element method.The influence of key constitutive parameters on the shape of the true stress-strain curve calculated by the polycrystalline uniaxial tensile model is summarized,providing a theoretical basis for determining the range of subsequent constitutive model parameters.Based on the morphological characteristics of polycrystalline uniaxial tensile true stress-strain curves,a prediction model for predicting the crystal plastic constitutive parameters of face-centered cubic and body-centered cubic crystal materials was established based on the BP neural network algorithm.The construction methods of input and output vectors,sample banks,and neural network structures for parameter prediction models are systematically discussed.In order to improve computational efficiency and save computational costs,this paper discusses the influence of hidden layer structure and sample number of neural networks on prediction results,providing a basis for selecting the optimal network structure and sample number.Finally,the reliability of predicting constitutive parameters is verified using quantitative indicators.Using 316L austenitic stainless steel with a face centered cubic structure and self developed ferritic stainless steel with a body centered cubic structure,the reliability of the two constitutive parameter prediction models for predicting actual material parameters was verified,respectively.Finally,the applicability of the prediction parameters at the macro and micro levels was discussed in combination with macroscopic uniaxial tensile experiments and nanoindentation experiments.