Machine Learning to Accelerate Full-Field Crystal Plasticity Simulations

Over the past century, a lot of research has been done to form innovative materials that can cater to serve different purposes in industries such as aerospace, automotive, marine etc. However, metals still hold a major part of the products developed in these industries and hence, it is important to study the behavior of metals both macroscopically as well as microscopically. Finite Element Method using Crystal Plasticity constitutive model (CPFEM) is one of the popular methods that enables investigating the material behavior at micro level accurately but is computationally expensive. Recent studies suggest that ML (Machine learning) techniques can significantly reduce the computational time when ensembled with methodologies that evaluate the behavior of materials (e.g., phase-field methods, CPFEM, Peridynamics). This paper highlights the ability of ML based full-field simulations of polycrystalline alloys to reduce computational time of CPFEM simulations. A convolution-based LSTM (ConvLSTM) model is used and trained using few time steps of full-field response from CPFEM simulations.