[SY-D5] A machine-learning approach for finding new hard-magnetic phases
Data-mining and machine-learning (ML) techniques play an increasingly important role in the discovery and development of new materials. In this contribution, we use kernel-based learning methods to predict optimal chemical compositions for new permanent magnets, which are key components in many green-energy technologies. The magnetic-property data used for training and testing the ML models were obtained by a combinatorial high-throughput screening (HTS) using density-functional theory calculations. For encoding the structural and chemical information of the HTS data in a machine-readable format, we use several existing and newly developed material descriptors and assess the predictive power of the ML models built with them. The accuracy of the ML models with an optimal choice of descriptor and model parameters enables the prediction of promising structurecomposition combinations for substitutes of state-of-the-art magnetic materials like Nd2Fe14B - with similar intrinsic hard-magnetic properties but no or less amounts of critical rare-earth elements.