A Machine learning perspective on hardness prediction in multicomponent Al-Mg based lightweight alloys

Abstract

Achieving the desired hardness in Al-Mg alloys through experimentation is challenging, costly, and time-consuming, given the extensive composition variations and diverse aging conditions. This research utilizes a range of machine learning (ML) techniques to expedite the advancement of high-performance Al-Mg alloys. The research focused on Al-Mg-X (where X represents elements like Cu, Zn, etc.) alloys, compiling data from literature sources that included composition of alloy, the conditions under which aging occurred (including both time and temperature factors), various physical properties, and the measurement of hardness. These datasets were employed to train seven different ML algorithms aimed at predicting Al alloys with enhanced hardnesses. The findings indicated that the CatBoost model proved effective in forecasting the hardness with excellent predictive performance, surpassing other machine learning approaches, as evidenced by various optimisation metrics.