Design of high strength and electrically conductive aluminium alloys by machine learning

Traditionally, thermodynamic modeling considers only the equilibrium conditions and one-dimensional evolution of phases. Therefore, it has difficulty in predicting the final properties of materials, especially when the functional and mechanical properties are correlated and heavily dependent on the combination of different phases which distribute in three dimensions. Recently, machine learning enabled us to establish the complex relationship between alloy compositions, processing conditions, various phases, and final properties. In this work, machine learning is coupled with thermodynamic calculations to optimise the alloy compositions, processing conditions, and the combinations of phases for improved electrical conductivity and mechanical property. Compared with previous chemistry design by machine learning for multiple inputs and single object outputs, the introduction of intermediate phases from thermodynamic calculations can improve the prediction accuracy. Combining machine learning with thermodynamic calculation is expected to optimise new alloys.

Automated summary

Traditional thermodynamic modeling has limitations in predicting properties of materials, especially when functional and mechanical properties are correlated and dependent on multiple phases in three dimensions. Machine learning coupled with thermodynamic calculations can optimize alloy designs and improve prediction accuracy for new alloys.