Azeotropy in the natural and synthetic refrigerant mixtures

A novel approach for the prediction of azeotrope formation in a mixture that does not require vapour–liquid equilibrium calculations is developed. The method employs neural networks and global phase diagram methodologies to correlate azeotropic data for binary mixtures based only on critical properties and acentric factor of the individual components in refrigerant mixtures. Analytical expressions to predict azeotropy and double azeotropy phenomena in terms of critical parameters of pure components and interaction parameters k12, are derived using global phase diagram conception. Modeling of thermodynamic and phase behavior has been carried out on the base of the Redlich–Kwong–Soave and the Peng–Robinson equations of state (EoS). Local mapping method is introduced to describe thermodynamically consistently an accurate saturation curve of refrigerants by three parameters EoS. Optimized neural network was chosen to achieve a complete coincidence of predicted and experimentally observable azeotropic states for training, validation, and test sets simultaneously. All possible cases of azeotropy appearance/absence in the more than 1500 industrially significant binary mixtures of natural and synthetic refrigerants are presented.