Machine learning-based mass density model for hard magnetic 14:2:1 phases using chemical composition-based features

TL;DR

This paper develops a machine learning model to predict the mass density of 14:2:1 magnetic phases based on chemical composition and lattice parameters, aiding in the design of high-performance permanent magnets.

Contribution

It introduces a novel ML-based density prediction model for 14:2:1 phases using composition and lattice features, with high accuracy on unseen data.

Findings

Achieved 0.51% MAE with composition-only features

Model effectively predicts densities for diverse phases

Provides a tool for accelerating magnetic material design

Abstract

The Fe14Nd2B-based permanent magnets are technologically sought-after for energy conversion due to their unparalleled high energy product (520 kJ/m3). For such 14:2:1 phases of different compositions, determining the magnetization from the measured magnetic moment is often bottlenecked by lack of mass density. We present a machine learning (ML) mass density model for 14:2:1 phases using chemical composition-based features (representing 33 elements) and optionally lattice parameter (LP) features. The datasets for training and testing contain 190 phases (177 compositionally different) with their literature reported densities and LP. With an ML model with merely compositional features, we achieved a low mean-absolute-error of 0.51% on an unseen test-dataset.