Aluminum Alloy Design by La Amount through Machine Learning and Experimental Verification

Abstract

The development and design of metal materials have been carried out through experimental method and simulation based on theoretic. Recently, with the widespread application of artificial intelligence (AI) in various fields, many studies have been actively incorporating artificial intelligence into the field of metal material design. Especially, many studies have been reported on adding rare-earth elements to aluminum alloys to improve corrosion resistance and mechanical properties using AI. However, the performance evaluation of artificial intelligence through experimental verification has not yet been reported related to metal material. In this study, we investigated the artificial intelligence algorithm capable of predicting the hardness based on the composition ratio of aluminum alloy with added Lanthanum (La) using experimental data and conducted a comparative analysis of the predicted hardness values. The machine learning models employed Adaptive Boosting Regressor (ADA), Gradient Boosting Regressor (GBR), Random Forest Regressor (RF), and Extra Trees Regressor (ET). The dataset comprised 1,210 encompassing 9 composition elements constituting the alloy. In the result, the findings revealed that the ET model demonstrated the most effective performance in predicting hardness. In addition, the microstructure became fine and showed the highest hardness at 0.5 wt.% La and hardness tended to decrease as the amount of La increased. The ET model showed excellent performance in predicting this tendency through experimental verification.