Abstract: In traditional trial-and-error method, enhancing the Young's modulus of magnesium alloys while maintaining a favorable ductility has consistently been a challenge. It is a need to explore more efficient and expedited methods to design magnesium alloys with high modulus and ductility. In this study, machine learning (ML) and assisted microstructure control methods are used to design high modulus magnesium alloys. Six key features that influence stiffness and ductility have been extracted in this ML model based on abundant data from literature sources. As a result, predictive models for Young's modulus and elongation are established, with errors <2.4 % and 4.5 % through XGBoost machine learning model, respectively. Within the given range of six features, the magnesium alloys can be fabricated with the Young's modulus exceeding 50 GPa and an elongation surpassing 6 %. As a validation, Mg-Al-Y alloys were experimentally prepared to meet the criteria of six features, achieving Young's modulus of 51.5 GPa, and the elongation of 7 %. Moreover, the SHapley Additive exPlanation (SHAP) is introduced to boost the model interpretability. This indicates that balancing the volume fraction of reinforcement, the most important feature, is key to achieve Mg-Al-Y alloys with high Young's modulus and favorable elongation through the two models. Enhancing reinforcement dispersion and reducing the size of reinforcement and grain can further improve the elongation of high-stiffness Mg alloy.

Key words: Image analysis, Machine learning, Al2Y, Young’s modulus