Mechanism-informed machine learning for intelligent design of rare-earth-containing magnesium-based alloys

Rare-earth (RE)-containing magnesium (Mg)-based alloys are promising for lightweight structures, electrochemical devices, and biodegradable implants. RE alloying regulates precipitation, grain structure, texture, interfacial chemistry, and degradation behavior, but also creates a highly coupled composition-processing-microstructure-property space. This review presents a mechanism-informed machine-learning (ML) framework for RE-containing Mg-based alloys, covering data integration, cross-scale feature engineering, model selection, uncertainty quantification, interpretability, target-oriented optimization, and experimental validation. Representative applications in mechanical properties, corrosion, electrochemical performance, and biodegradation are critically compared. Current studies have progressed from single-property prediction toward candidate screening and multi-objective optimization. However, strict inverse design remains limited by scarce and heterogeneous data, incomplete physical descriptors, weak model transferability, and insufficient closed-loop validation. Future progress requires standardized Mg-specific databases, physics-informed models, uncertainty-guided active learning, and mechanism-constrained optimization.