Data-driven electrochemical behavior prediction for refractory high-entropy alloys by global and focused learning

Abstract

Curve data are essential tools in materials science for characterizing material properties. However, obtaining and analyzing these curve data such as electrochemical corrosion curves to establish the intrinsic relationship of the material is time-consuming work. While machine learning (ML) method can dramatically accelerate material research and development, accurately predicting electrochemical curves to understand the corrosion behavior of corrosion-resistant alloys remains a significant challenge, due to that macroscopic experiments and microscopic theoretical simulations have yet to be effectively integrated. In this work, we propose a data-driven method that integrates global and focused learning (GFL) strategies. Taking refractory high-entropy alloys (RHEAs) as a case study, we establish prediction models for their corrosion behavior based on potentiodynamic polarization curve data and interpretable GFL. Through compositional optimization, a series of RHEAs with high corrosion resistance, such as Ti₂₀V₁₀Nb₂₀Mo₂₀Ta₃₀, have been obtained. This alloy exhibits excellent corrosion resistance compared with other RHEAs. In addition, compared with traditional single ML methods, GFL not only accurately predicted the polarization curves of RHEAs but also captures the key factors affecting the corrosion resistance of the alloys. The GFL strategy provides an effective ML tool with physical interpretation for material curve data analyzation.