Active learning prediction and electromagnetic levitation measurement of metastable liquid properties of Ti-Al-Zr alloy

The accurate knowledge of liquid properties for Ti-Al based alloys is of great significance for scientific explorations such as revealing the atomic-scale behaviors in alloy processing and realizing the optimization of manufacture technology. By means of the active learning method based on deep neural network (DNN) and electromagnetic levitation technology (EML), we investigate the microstructure evolution of Ti 47 Al 47 Zr 6 alloy, and with a focus on obtaining its liquid state properties. The density, surface tension, viscosity of this liquid alloy and the related self-diffusion coefficient are predicted by the DNN potential. The calculated surface tension exhibits only a deviation of less than 2 ​% as compared with EML experimental values. The liquid local structure characteristics acquired through Voronoi polyhedron analysis indicate that the fraction of relatively high-coordinated clusters with Al as the central atom displayed an anomalous decrease with the falling of liquid temperature. This effect is attributed to the tendency of these clusters to form icosahedral-like geometries, resulting in an increased fraction of icosahedral-like geometries in liquid alloy.