Accelerating age hardening behavior of Mg alloy using a data-driven screening strategy guided by machine learning

In this work, we introduce a self-guided design framework that combines target property and feature importance evaluation to optimize the aging behavior of Mg alloys. The design efficiency was built by a precise and reliable two-step feature engineering training strategy (2SFE) guided by machine learning models capable of assessing a range of Mg alloy properties across high-dimensional datasets. A multi-objective hybrid algorithm combines efficient global optimization (EGO) and a gradient-based feature reduction is established within this framework. EGO incorporates an innovative multi-objective gradient designed to speed up the search process along the Pareto front scheme, while 2SFE reduces features to improve the design space. A novel two-step feature engineering selection strategy (2SFE) was introduced, reducing computational input features to six (6) while enhancing prediction accuracy. Compared to the traditional approach, the scaled super learner model improved accuracy from 0.86 to 0.97. This methodology was experimentally validated, resulting in the creation of a novel Mg-6Gd-Zr-6Y-Nd, Mg-9Zn-5Gd-Zr-8Y-3Nd, Mg-8Zn-Zr-3Y-Nd, and Mg-9Zn-Zr-2Y (wt%) alloy, processed under defined specific conditions. The integrated approach was validated and assessed by (a) comparing predicted values with experimental results and (b) benchmarking the model’s outcomes with existing experimental and theoretical findings reported for Mg alloys. This integrated approach offers a cost-effective, efficient pathway for designing advanced Mg alloys while significantly reducing manual input in the alloy development process.