Accelerated discovery of Fe-based amorphous/nanocrystalline alloy through explicit expression and interpretable information based on machine learning

Abstract

The intricate interplay between the characteristics and properties of amorphous/nanocrystalline alloys, specifically saturation flux density (Bs) and Curie temperature (Tc), has long been a perplexing task, despite their remarkable properties. However, recent developments in the field of machine learning (ML) have offered a promising paradigm for the accelerated discovery of these elusive alloys. Notably, it has become clear that their properties depend not only on their chemical composition but also on their thickness and annealing process. To further advance this paradigm, 5 ML methods were employed to predict Bs and Tc based on original features (OFs) and polynomial features (PFs). Impressively, it was discovered that ML model with the largest R2 score on OFs displayed an outstanding capability, with interpretable method, which could be of great aid to alloy design. Further analysis revealed that the explicit formulas based on PFs quantitatively provided direction for optimization. Finally, the Fe81.4Si3.4B11.4Cu0.8Nb3 alloy, with appropriate annealing time (AT1), annealing temperature (AT2) and thickness (THK), demonstrated promising Bs = 1.79 T and Tc = 661.6 K. This work strongly suggests that the discovery paradigm has the potential to quantitatively explore the relationship between features and properties and might provide some insight into the discovery of new Fe-based amorphous/nanocrystalline alloy.

Keywords

• Amorphous/nanocrystalline alloy
• Soft magnetic property
• Machine learning
• Interpretation
• Polynomial feature