Machine learning-assisted design of refractory high-entropy alloys with targeted yield strength and fracture strain

Abstract

In order to improve the traditional “trial and error” material design method, machine learning-yield strength and machine learning-fracture strain models are incorporated into one system to predict yield strength and fracture strain in refractory high-entropy alloys (RHEAs) under compression. The ML-yield strength model and ML-fracture strain model achieve excellent predictions (R2 = 0.942, RMSE=0.35) and (R2 = 0.892, RMSE=0.41) in the testing set, respectively. Based on the machine learning model, Nb0.22Ta0.22Ti0.24V0.23W0.09, Nb0.24Ta0.22Ti0.26V0.04W0.24, Nb0.26Ta0.24Ti0.21V0.24W0.05, and Nb0.18Ta0.26Ti0.22V0.21W0.13 RHEAs in the Nb-Ta-Ti-V-W RHEA system were screened and synthesized. The yield strength (1915 MPa, 1983 MPa) of the Nb0.22Ta0.22Ti0.24V0.23W0.09 and Nb0.24Ta0.22Ti0.26V0.04W0.24 RHEAs are higher than that (1689 MPa) of the NbTaTiVW RHEA. The unfractured Nb0.18Ta0.26Ti0.22V0.21W0.13 and Nb0.26Ta0.24Ti0.21V0.24W0.05 RHEAs under compression exhibit superior performance than the fracture strain (16.6 %) of the NbTaTiVW RHEA. The mixing enthalpy of RHEAs is negatively correlated with the yield strength, whereas a negative relationship exists between electronegativity difference and fracture strain through the SHAP analysis. Decreasing the mixing enthalpy and increasing the electronegativity difference promote the formation of the precipitated phase. The electron probe microanalysis reveals that the differences in mechanical properties (yield strength and fracture strain) in the NbTaTiVW RHEAs primarily stem from the fraction of the precipitated phase.