Title: Artificial Intelligence Across Disciplines: Research, Innovation, and Intelligent Solutions
Editors: Dr. Subita Bhagat, Dr. A. Balamurugan, Dr. P. Krishna Kumar, and Mrs. S. Nandhini Devi
ISBN: 978-81-69857-83-3
Chapter: 22
DOI: https://doi.org/10.59646/815/22
Author: Dr. V. Kathiravan
Abstract
Machine learning (ML) has fundamentally restructured computational materials science, transitioning materials discovery from empirical trial-and-error paradigms to predictive, data-driven inverse design frameworks. This chapter establishes an integrated predictive and generative framework for discovering and optimizing refractory high-entropy alloys (RHEAs) and advanced functional crystalline systems operating in extreme environments. Using a curated corpus of multi-component alloy compositions and high-throughput ab initio calculations, we deploy a hierarchical computational architecture combining physical thermodynamic feature engineering, gradient-boosted decision trees, crystal graph convolutional neural networks (CGCNN), and Bayesian active learning. The model achieves an -score of 0.941 in solid-solution phase classification and lowers continuous yield strength regression error to 42.6 MPa (a 68.2% error reduction relative to empirical Hume-Rothery criteria). By guiding closed-loop vacuum arc melting and high-temperature nanoindentation, the framework discovers the single-phase body-centered cubic alloy
, exhibiting an ambient yield strength of 1,485 MPa and high-temperature strength retention exceeding 820 MPa at 1,000°C.
Keywords: Materials Informatics, High-Entropy Alloys, Crystal Graph Convolutional Neural Networks, Active Learning, Phase Stability, Density Functional Theory.