Microstructural Characterization and Phase Prediction of High Entropy Alloys using Empirical and Machine Learning Based Methodologies

Abstract

Alloys containing four or more principal alloying elements in near equal atomic percentage, so called High Entropy Alloys (HEA), break down the solvent-solute relationship generally seen in traditional alloys. This creates a vast and unpredictable design space, unfeasible to explore purely experimentally or through computational simulations. This thesis aims to present a newly developed machine learning based phase prediction methodology for HEA alloys, the empirical design and characterization of a Zr-containing HEA system, and a performance comparison of the developed methodology and established empirical method at predicting the impact of Zr on the phase composition with experimental results as a reference. Findings show that the developed phase prediction methodology, which leverages a convolutional neural network, could predict the formation of a secondary BCC solid solution phase within the high entropy alloys caused by the addition Zr and intermetallic compound formation, while the empirical parameters failed to do so.