Prediction of Solvent Activity in Polymer Systems with Neural Networks

Abstract

It is an arduous task to develop thermodynamic models or empirical equations which accurately predict solvent activities in polymer solutions. Even so, since Flory developed the well-known equation of state for polymer solutions, much work has been carried out in this area. Consequently, extensive experimental data have been published in the literature by various researchers on different polymer binary systems. When such data are available, then modeling solvent activity in polymer solutions can be simplified by the use of the artificial neural network (ANN) technique. The neural network technique has been in existence since 1969, when Grossberg introduced a network algorithm that could learn, remember, and reproduce any number of complicated space-time patterns. Nevertheless, the use of ANN to predict thermodynamic and fluid properties is still rather limited. In this paper, an attempt has been made to predict the activity of several polymers in different solvents. We present a simple feed-forward neural network architecture with a nonlinear optimization training routine for this purpose. The predictions of the proposed training routine were then compared with other traditional training routines such as the error-back-propagation and Madaline III. The predictions generated by all three algorithms were good, but the proposed algorithm was much faster and yielded better results.