Timothy Liao (The University of Texas at Austin)

"The discovery and design of rare-earth free magnets using machine learning"

Abstract: What do wind turbines, electric car motors, and computer hard disks have in common? They all rotate, yes, but they also contain powerful magnets.

The high-performance permanent magnets found in these devices contain rare-earth elements like neodymium or samarium. Yet, given the supply risks associated with rare-earth elements, my dissertation research focuses on discovering strong magnets without them. I leverage efficient first principles calculations and machine learning to accelerate the typically slow trial-and-error method in experiments. This presentation outlines my team's efforts, culminating in the discovery of a new material, Fe3CoB2, with significant implications for green energy and advanced technology.

Initially, I establish a magnetic materials database encompassing magnetization (i.e., magnetic strength), magnetic anisotropy (i.e., magnetic field resilience), and Curie temperature (i.e., heat resilience). Then, I enhance the functionality of a popular machine learning (ML) model, Crystal Graph Convolutional Neural Network, to predict both macroscopic and microscopic properties, validated through high-throughput first-principles calculations on Fe-Co-N.

Following successful ML model testing, I introduce an adaptive machine learning feedback system for efficient materials discovery. Our framework streamlines the computational discovery and experimental synthesis of Fe3CoB2, reducing the process to mere days. Building on this achievement, I extend the search to Fe-Co-Si compounds, expected to be compatible with a Si substrate. Finally, I evaluate the performance of our top candidate Fe-Co-X (X = N, B, Si, P, and C) structures against common rare-earth-free magnets.