Harnessing machine learning for predicting mechanical properties of lightweight Mg alloys

Abstract

In this study, we introduce a machine learning methodology for predicting mechanical properties in Mg-based multicomponent alloys, incorporating alloying elements and processing methods as input variables. We employed eight ML models and assessed their efficacy using metrics such as R2, RMSE, and MAE after using five-fold cross validation grid search hyperparameters tuning process. Following this, we implemented the top-performing Extra Tree (ET), Random Forest (RF) and XGB models to predict mechanical properties for Mg alloys. The best performance was observed with an R2 of 95.1 % and 97.2 %, RMSE of 7 % and 8 %, and MAE of 4.7 % and 5 % for UTS and YS respectively using the Extra Tree model. This research not only showcases the efficiency of ML techniques with minimal intervention but also offers valuable insights paving the way for accelerated design of Mg-alloys for tailored application in the future. © 2024 Elsevier B.V.