Design of Ni-based single crystal superalloys by machine learning based on data-driven multi-task optimization

Abstract

In order to ascertain the requirements of high-quality gas turbine components, an investigation was undertaken into the composition design of Ni-based single crystal superalloys utilising high-throughput, data-driven machine learning methodologies. A characteristic parameter space influencing high-temperature strength in superalloys was established through theoretical calculation and the extraction of data from the superalloy manual. By means of correlation screening, it was determined that Vγ′、V_TCP and Tγ′-solvus represent the crucial parameters, which were used as the basis for optimisation tasks. In accordance with the principles of domain knowledge, a series of constraints were established for the purpose of screening the prospective alloy components, with the objective of achieving a high level of temperature strength. Subsequently, the preferred composition alloys and the commercial alloys were prepared through experimentation. Characterisation of the new composition alloys revealed that they exhibited a microstructure with 60 ∼ 75 % cubic γ′ phase, which demonstrated a relatively strong correlation with the optimisation tasks. Furthermore, the results show that the new alloys exhibit not only superior high-temperature strength but also notable improvements in elongation, density and cost, conferring them with enhanced economic value and potential for wider application compared to the current commercial alloys. The correlation between microstructure and properties suggests that the enhancement in performance is attributable to the optimisation of key microstructure objectives, offering a promising avenue for alloy design.