By integrating published experimental data on the uranium-zirconium (U-Zr) system into a machine learning framework, insight into the two differing views on the thermochemical equilibrium, particularly on the U-rich portion of the phase diagram (PD) was developed, ultimately resulting in a new U-Zr PD. Phase diagram sensitivity to model parameters, tolerances, physical preconceptions and experimental biases, are considered to establish the validity of the generated PDs. A systematic assessment of the most reliable and most recent thermochemical data was made, and the traditional modeling bias to search the space of free energy parameters was removed by using recently developed machine learning strategies. The readily validated methodology enables a thermodynamically consistent search of free energy parameters by leveraging modern experimental work from an array of sources including phase transformations, phase transition temperatures, and enthalpy changes between 723-1173 K (450-900°C). These changes include the truncation of β-U stability at 6 at.% Zr, prominent isotherms at 884 K (611°C) and 961 K (688°C), and δ-U-Zr phase boundaries ranging from 66.5 to 80.2 at.% Zr at 823 K (550°C). The newly proposed PD captures fundamental constants measured experimentally and improves the agreement with phase transformation studies such as neutron diffraction with in situ heating. As such, it is proposed that the new U-Zr PD developed in this work be used to resolve the historically opposing views.
Read full abstract