In this paper, an efficient design of a Ti-modified Al-Si-Mg-Sr casting alloy with simultaneously enhanced strength and ductility was achieved by integrating computational thermodynamics, machine learning, and key experiments within the Bayesian optimization framework. Firstly, a self-consistent Al-Si-Mg-Sr-Ti quinary thermodynamic database was established by the calculation of phase diagram method and verified by key experiments. Based on the established thermodynamic database, a high-throughput Scheil-Gulliver solidification simulation of the A356-0.005Sr alloy with different Ti contents was carried out to establish the "composition-microstructure" quantitative relationship of the alloy. Then, by combining the computational thermodynamic, machine learning, and experimental data within the Bayesian optimization framework, the relationship "composition/processing-microstructure-properties" of A356-0.005Sr with different Ti contents was constructed and validated by the key experiments. Furthermore, the optimum alloy composition of the Ti-modified A356-0.005Sr casting alloy was designed based on this integration method with the Bayesian optimization framework and verified by the experiments. It is anticipated that the present integration method may serve as a general one for the efficient design of casting alloys, especially in the high-dimensional composition space.
Keywords: CALPHAD; Ti modification; alloy design; casting aluminum alloy; machine learning.