The viscosity surfaces of R152a in the form of multilayer feed forward neural networks

Abstract

A multilayer feedforward neural network (MLFN) technique is adopted for developing a viscosity equation η=ηρ,T for R152a. The results obtained are very promising, with an average absolute deviation (AAD) of 0.36% for the currently available 300 primary data points, and they are a significant improvement over those of a corresponding conventional equation in the literature. The method requires a high accuracy equation of state for the fluid in order to convert the experimental P,T into the independent variables ρ,T, but such equation may not be available for the target fluid. Aiming at overcoming this difficulty, two viscosity explicit equations in the form η=ηP,T, avoiding the density variable, are also developed, one for the liquid surface and the other for the vapor one. The reached accuracy levels are equivalent to that of the former η=ηρ,T equation. The trend of the reduced second viscosity virial coefficient is correctly reproduced in the data range. The proposed technique, being heuristic and non theoretically founded, is also a powerful tool for experimental data screening.