Machine learning-driven phase prediction and corrosion behavior of (CoCrNi)(100-x-y) AlxTiy high-entropy alloys in Ringer's solution

Xin Zhao, Mengdi Zhang, Hanqing Xu, Zhuoyi Wang, Tianming Li, Rui Li,
Machine learning-driven phase prediction and corrosion behavior of (CoCrNi)(100-x-y) AlxTiy high-entropy alloys in Ringer's solution,
Materialia,
Volume 44,
2025,
102590,
ISSN 2589-1529,
https://doi.org/10.1016/j.mtla.2025.102590.
(https://www.sciencedirect.com/science/article/pii/S2589152925002583)
Abstract: Limited by traditional trial-and-error methods, it is a great challenge to develop novel high-entropy alloys (HEAs) with an FCC+BCC dual-phase structure and excellent corrosion resistance. Herein, this study developed a machine learning (ML)-based design method, which predicted the influence of Al-Ti co-doping on the phase structure of CoCrNi-based HEAs and used this as a screening criterion to obtain the target alloys. After model optimization and comparative evaluation, the Random Forest (RF) algorithm was ultimately selected for phase prediction, achieving an accuracy of 94.1 %. To verify the accuracy of the machine learning phase prediction model, two types of HEAs were designed: one is composed of (CoCrNi)94Al3Ti3, (CoCrNi)94Al4Ti2, and (CoCrNi)93Al4Ti3 with a single FCC structure, and the other comprises (CoCrNi)90Al5Ti5, (CoCrNi)85Al8Ti7, and (CoCrNi)80Al10Ti10 with an FCC+BCC dual-phase structure. SHAP analysis was employed to enhance the interpretability of the model, and the results showed that valence electron concentration (VEC) exerts the most significant influence on phase formation. In addition, electrochemical test results of the FCC+BCC dual-phase HEAs in Ringer's solution indicated that the Al5Ti5 alloy exhibits the optimal corrosion resistance, with a corrosion current density of 8.08×10⁻⁸ A/cm², a pitting potential of 840.6 mV, and a passive region of 1062.4 mV.
Keywords: High-entropy alloys; Machine learning; Shap analysis; Phase; Corrosion behavior; Passive film