Machine learning assisted design of Ni-based superalloys with excellent high-temperature performance

Due to compositional complexity, designing superalloys with multiple targeted properties is a great challenge. Here, we propose a workflow that incorporates composition and high-temperature mechanical performance into machine learning to optimize commercial K403 superalloys for the needs of increasing service temperature of hot-end parts in aircraft engine and gas turbine. Moreover, multiple properties including microstructure stability, the volume fraction of γ’ precipitates, processing window, freezing range and density were simultaneously optimized to select 7 superalloys from 15,625 candidates. One selected superalloy was experimentally synthesized. Compared with the commercial K403 superalloy, the creep rupture life of the newly-designed superalloy is improved around three times at 975 °C and even 1025 °C. The predicted high-temperature creep rupture life and yield strength using the machine learning model is in excellent agreement with experiments. The current machine learning approach provides guidance for the rapid design of multi-component superalloys with targeted multiple desired functionalities.