Viscosity correlation for ethane in the form of multilayer feedforward neural networks

A multilayer feedforward neural network (MLFN) technique is proposed for developing viscosity correlations. The accuracy and usefulness of the method is demonstrated by developing the viscosity correlation for ethane as a function of temperature and density and comparing its predictive power to that of the traditional ethane correlation. The overall average absolute deviation (AAD), based on the primary data set, is the same for both correlations, while the bias of the MLFN correlation is much lower. It has been successfully demonstrated, by examining the viscosity of liquid ethane, that the MLFN technique is capable of fitting the viscosity as a function of temperature and pressure, thus bypassing the need for an accurate equation of state. The resulting MLFN correlation for the viscosity of liquid ethane reproduced the experimental data with the same accuracy as the traditional correlation with temperature and density as dependent variables. The excellent agreement obtained indicates that MLFN should be the preferred method of choice, especially when developing viscosity correlations in terms of temperature and pressure.