Graph Neural Networks to Predict Coercivity of Hard Magnetic Microstructures

TL;DR

This paper demonstrates how graph neural networks can effectively predict coercivity and energy product in hard magnetic microstructures, aiding the discovery of rare-earth free permanent magnets.

Contribution

It introduces a GNN-based approach trained on magnetic simulation data to predict coercivity and energy product, including uncertainty quantification and out-of-distribution analysis.

Findings

GNN accurately predicts coercivity with quantified uncertainty.

The model generalizes to out-of-distribution microstructures.

Feature engineering improves prediction of coercivity dependence on system size.

Abstract

Graph neural networks (GNN) are a promising tool to predict magnetic properties of large multi-grain structures, which can speed up the search for rare-earth free permanent magnets. In this paper, we use our magnetic simulation data to train a GNN to predict coercivity of hard magnetic microstructures. We evaluate the performance of the trained GNN and quantify its uncertainty. Subsequently, we reuse the GNN architecture for predicting the maximum energy product. Out-of-distribution predictions of coercivity are also performed, following feature engineering based on the observed dependence of coercivity on system size.