Data-driven design of corrosion-resistant complex concentrated alloys

Abstract

Complex concentrated alloys (CCAs) represent an emerging class of multi-principal element materials with exceptional mechanical and thermal stability. However, designing corrosion-resistant CCAs remains a significant challenge due to the complex interplay of thermodynamic, electronic, and environmental factors. In this work, a data-driven framework integrating machine learning (ML) and Bayesian optimization (BO) was employed to identify CCA with superior corrosion resistance. Eight machine-learning models were evaluated, and the eXtreme gradient boosting algorithm provided the highest predictive accuracy for corrosion rate. A three-stage feature screening strategy identified mixing enthalpy (ΔHmix), geometrical parameter (λ), and valence electron concentration as the most influential descriptors, with ΔHmix calculated using the CALPHAD method to accurately capture atomic interaction effects. BO based on these descriptors yielded the optimal composition Co15.34Cr23.47Fe17.02Ni 26.21Ti8.36Mo9.61, which was synthesized and experimentally validated. Potentiodynamic polarization confirmed exceptional corrosion resistance, including a corrosion current density on the order of 10−9 A cm−2 and a corrosion rate of approximately 9.5 × 10−5 mm yr−1. X-ray diffraction revealed a dominant BCC phase with a less amount of FCC phase. Strong agreement between predictions and experiments validates the ML based optimization framework for rapid design of corrosion-resistant CCAs.