Developing novel low-density high-entropy superalloys with high strength and superior creep resistance guided by automated machine learning

Design of novel superalloys with low density, high strength, and great microstructural stability is a big challenge. This work used an automated machine learning (ML) model to explore high-entropy superalloys (HESAs) with coherent γ' nanoprecipitates in the FCC-γ matrix. The database samples were firstly preprocessed via the domain-knowledge before ML. Both autogluon and genetic algorithm methods were applied to establish the relationship between the alloy composition and yield strength and to deal with the optimization problem in ML. Thus, the ML model cannot only predict the strength with a high accuracy (R² > 95 %), but also design compositions efficiently with desired property in multi-component systems. Novel HESAs with targeted strengths and densities were predicted by ML and then validated by a series of experiments. It is found that the experimental results are well consistent with the predicted properties, as evidenced by the fact that the designed Ni-5.82Fe-15.34Co-2.53Al-2.99Ti-2.90Nb-15.97Cr-2.50Mo (wt.%) HESA has a yield strength of 1346 MPa at room temperature and 1061 MPa at 1023 K and a density of 7.98 g/cm³. Moreover, it exhibits superior creep resistance with a rupture lifetime of 149 h under 480 MPa at 1023 K, outperforming most conventional wrought superalloys. Additionally, the coarsening rate of γ' nanoprecipitates in these alloys is extremely slow at 1023 K, showing a prominent microstructural stability. The strengthening and deformation mechanisms were further discussed. This framework provides a new pathway to realize the property-oriented composition design for high-performance complex alloys via ML.