Machine learning prediction and experimental validation of magnetic properties of M-type hexagonal ferrites

Abstract

M-type hexagonal ferrites have received widespread attention for their excellent magnetic properties, but studying the intricate interplay between their composition, process and magnetic properties has long been a puzzling task. The advent of machine learning offers a promising example for accelerating the discovery of connections between them. In this work, we trained five machine learning regression models (SVR, DTR, AdaBoost, GBDT and MLP) using data collected from the literature and found that the SVR model performed well on the dataset. In order to verify the feasibility of the SVR model, we predicted the magnetic properties of Ba1-xLaxFe12-yMnyO19 (0.1≤x=y≤0.6), and then prepared and tested them. The results showed a good agreement between the prediction and the experiment, which proved the reliability of the designed model. Moreover, we elucidate the occupancy trend of Mn3+ ions by first principles calculation and Raman spectroscopy verification. This work strongly suggests that machine learning has the potential to explore the relationship between composition, process and properties, which opens the door for the tailoring the properties of M-type hexagonal ferrites.