| Abstract Scope |
Historically, the CALculation of PHAse Diagrams (CALPHAD) has relied on trial-and-error model selection and optimization by using, sometimes incomplete, experimental and ab initio data as a means to predict a physics-based free energy description of the material’s thermochemical equilibrium. Although CALPHAD has greatly advanced our understanding of materials and their properties, the iterative process involves the fitting of coefficients for each phase, individually and in pairs, leading to cumbersome and time-consuming efforts across months and years that obfuscate the development of a meaningful understanding of the underlying physics of the system at hand. In this context, a complete portfolio of integrated physics-informed machine learning tools is presented to predict and optimize arbitrary Gibbs free energy models, in 3 to 30 minutes on a single desktop computer. The generalized architecture is demonstrated for over 50 material systems, including example metals, ceramics, and polymers, based solely on their experimental phase diagram. |