Design of high strength and ductile recycled Al alloys by machine learning

Recycled Al alloys suffer from brittle Fe-rich intermetallics due to the Fe impurities accumulated during the recycling process and degrading mechanical properties such as low strength and poor ductility. In this study, machine learning methods such as BP neural network (BP), Extreme Learning Machine (ELM), Support Vector Machine (SVM) and Random Forest (RF) have been used to establish a regression model for predicting the strength and elongation of the recycled Al alloys under different input features. Further, the effects of alloy composition, secondary phase percentage, and heat treatment process on the prediction accuracy of strength and elongation has been investigated, and the highest prediction accuracy of the machine learning models are determined with the coupled input features of alloy composition and secondary phase percentage, while the RF model demonstrated the superior prediction performance both the strength and elongation regression models. Based on the significance analysis of the input features of the RF model, the elements of Fe, Y and La as well as the Fe-rich, Y-rich and La-rich phases, have the greatest impact on the strength and ductility of the alloys. Finally, the mechanical properties of the recycled Al alloys with different rare earth (RE) additions (0–0.2 %) were predicted, and the optimum addition of RE elements was determined to be 0.10 wt% Sc, which was verified by the X-ray CT quantification of Fe-rich phases and microporosity, achieving an improvement in tensile strength by 4.37 %. This study provides a guide for designing the composition of Fe-bearing recycled Al alloys by machine learning.