Machine learning-driven design of rare metal doped niobium alloys with enhanced strength and ductility

Abstract

The doping of rare metal elements (RMEs) in niobium alloys offers significant potential for improving mechanical properties by refining grains and forming secondary phase particles. However, the inherent trade-off between strength and ductility, coupled with the complexity of alloy composition-performance relationships, limits the efficiency of traditional trial-and-error methods. In this study, a machine learning-based framework was developed to optimize the mechanical properties of niobium alloys. A comprehensive database of niobium alloys' properties was analyzed using feature engineering, and a high-accuracy prediction model, Gray Wolf Optimization-Extreme Learning Machine (GWO-ELM), was constructed, achieving R2 values of 0.95 and 0.88 for tensile strength and elongation, respectively. The model was integrated with the Non-dominated Sorting Genetic Algorithm (NSGA-III) to design alloys with superior comprehensive properties. The optimized Nb521–0.116Sc alloy demonstrated a tensile strength of 772 MPa and an elongation of 10.3 %, with a maximum K value (representing comprehensive performance index) of 13.80, representing a 34.82 % improvement in the comprehensive performance index. Microstructural analysis revealed that the enhancements were primarily due to solid solution strengthening with high solubility. This study highlights the potential of machine learning in accelerating the design of high-performance niobium alloys and provides a robust strategy for developing advanced materials with balanced strength and ductility.

Keywords

• Niobium alloys
• Machine learning
• Solid solution
• Rare metal elements
• Mechanical properties