Data-driven multiobjective optimization of oxidation resistance and phase stability in refractory high-entropy alloys

Refractory high-entropy alloys and refractory complex concentrated alloys are considered promising candidates for ultra-high-temperature structural applications, yet their broader implementation is limited by inadequate oxidation resistance and phase instability under severe thermal conditions. In this work, a data-driven, multi-objective machine-learning framework is introduced to simultaneously evaluate oxidation resistance and phase stability in refractory alloy systems. A comprehensive dataset was constructed from published high-temperature oxidation studies, incorporating alloy compositions, oxidation parameters, and experimentally reported specific mass gain values. Oxidation resistance was characterized using specific mass gain, while phase stability was represented by a semi-quantitative indicator derived from reported microstructural observations. Ensemble-based regression models were trained and validated using k-fold cross-validation, demonstrating reliable predictive performance despite the heterogeneity of literature-derived data. Feature importance analysis reveals that oxidation behavior is predominantly controlled by aluminum and chromium, whereas phase stability is strongly influenced by refractory elements, particularly niobium. Pareto-based multi-objective optimization identifies a set of alloy compositions that balance low oxidation rates with favorable phase stability, confirming that the proposed framework captures physically meaningful trends beyond purely statistical correlations.

Перевод пока недоступен