Machine Learning Vacancy Formation Energy in Nickel-Based Superalloys

Creep performance plays a key role in nickel-based superalloys for high temperature applications. Creep behavior depends on many parameters such as strength, dislocations, diffusivity, and microstructural stability in addition to temperature, applied stress, and oxidation. This work focuses on predicting vacancy formation energy in nickel-based superalloys using machine learning approach. It is believed that vacancy formation energy and vacancy concentration will directly impact diffusion and dislocation behavior hence creep performance. High throughput density functional theory (DFT) calculations are performed on Ni-based alloys with addition of various alloying elements to predict the vacancy formation energy and vacancy concentration. Reported literature data on vacancy formation energy are also collected and evaluated to supplement the present DFT calculations. Machine learning is performed using various models including Atomistic Line Graph Neural Network. In light of the machine learning results, important features that have high correlation to vacancy formation will be discussed.