Multi-objective optimization of Fe-based metallic glasses on magnetic property, thermal stability and glass forming ability

Abstract

It is a great challenge to optimize the composition and improve the multi-properties of Fe-based metallic glasses (Fe-based MGs) by traditional trial-and-error method. Here, a strategy combining ensemble learning (EL) and non-dominated sorting genetic algorithm-III (NSGA-III) is proposed to efficiently design Fe-based MGs with simultaneous high magnetic property, thermal stability and glass-forming ability (GFA). With alloy compositions as the sole input, four EL models and three linear models are employed to predict saturated magnetic flux density ($B_s$), onset crystallization temperature ($T_x$) and maximum casting diameter ($D_{\max }$) of Fe-based MGs, respectively. Particle swarm optimization (PSO) algorithm is introduced to optimize hyperparameters of the models. It is found that the PSO-CatBoost model and PSO-ET models with different hyperparameters can better predict $B_s$, $T_x$ and $D_{\max }$ with higher accuracy. Further, the effects of key compositions on three properties are qualitatively analyzed based on SHAP analysis and correlation analysis. Finally, taking Fe–B–Si–P–C system as an example, multi-objective optimization based on NSGA-III is executed to obtain Pareto front and 15 Pareto-optimal solutions for three properties. When the $B_s$ is greater than 1.6 T, there are four alloys with $T_x$ and $D_{\max }$ better than the raw data. At this time, $T_x$ increases by 5.02% to 8.07% and $D_{\max }$ increases by 52.38% to 72.08%. It is also demonstrated that the sum of Fe and B content is the key factor affecting these three properties. This work provides an efficient strategy for developing new Fe-based MGs with better comprehensive performance.