Machine learning-based accelerated property prediction of two-phase materials using microstructural descriptors and finite element analysis

This study explores the use of supervised machine learning (ML) to predict the mechanical properties of a family of two-phase materials using their microstructural images. Random two-phase microstructures with a diversity of inclusion volume fractions, size distributions, and/or shapes are input into a finite element analysis program to determine the elastic modulus, Poisson's ratio, and phase stresses. The finite element analysis results establish the “ground truth” to train the supervised ML models. Two-point correlation (TPC) functions and principal component (PC) analysis are applied to the microstructures before training and testing the artificial neural network (ANN) and forest ensemble MLs. The chosen ML methods are found to accurately predict the homogenized elastic properties. Although the PCs for each set of microstructures are unique, recognizable patterns are detected that signify microstructural features that are key to microstructure-based property prediction. This work enables the development of ML algorithms to predict the mechanical properties of complex, multi-phase microstructural composites, based on microstructural images.