Correlation of the Transport Properties for the Ethanol-Water System Using Neural Networks

Abstract

Process design and simulation rely heavily on the accuracy and availability of transport property correlations. General models that combine the properties of pure components often lack the necessary accuracy. In this investigation, neural networks were used to model some important transport properties for the ethanol-water binary system. Specifically, a three-layer feed-forward neural network with six neurons in the hidden layer was used to model viscosity, thermal conductivity, surface tension and the Fick diffusion coefficient based on an array of experimental data. These neural network models were then compared to some conventional models that are commonly used to predict the aforementioned transport properties. The results showed that the neural network models were able to represent the experimental data very well for the system studied. One advantage in using neural network models to represent these properties is their ability to predict complex and interrelated behaviors without a priori information about the underlying model structure. Further, since all the models retain the same simple matrix structure, their integration into computer codes becomes straightforward and non-repetitive.