Age hardening stands as a crucial strengthening process for aluminium (Al) alloys. However, determining the optimal ageing conditions has traditionally relied on resource-intensive trial-and-error methods. To streamline the design of Al alloys, this study introduces a novel feature screening-assisted machine learning (ML) approach to explore age-hardening behaviour across varying compositions and heat treatment parameters, focusing on yield tensile strength (YTS), ultimate tensile strength (UTS), and elongation (EL). A comprehensive pool of features, incorporating alloy composition and fundamental elemental properties, was generated to provide metallurgical insights for ML predictions. Subsequently, feature screening methodologies, including correlation analysis, feature elimination, and multi-objective genetic algorithm (MOGA), were employed to identify key features from the vast feature pool, balancing model complexity and prediction accuracy. Employing support vector regression models trained on these optimized feature sets, we demonstrate enhanced prediction accuracy for strength properties while maintaining model simplicity for EL, with minimal impact on accuracy. In addition, experimental results further validate the method's ability to reproduce ageing curves across all three mechanical properties for both commercialized and newly designed Al alloys.
Read full abstract