Modelling of Vapour Liquid Equilibrium by Artificial Neural Networks

Vapour liquid equilibrium is condition wherein the liquid and vapour state of the components of a system are in equilibrium with each other. Conventionally, the vapour liquid equilibrium data is evaluated using the thermodynamic models, namely the equation of state (EOS), and the activity co- efficient models. The models falling under these categories are Peng-Robinson model, Margules model, vaanLaar model, Wilson's model, NRTL, UNIQUAC and UNIFAC model. VLE data is required in designing distillation columns and any doubt or inaccuracy in the prediction of the VLE data results in variation in design parameters which leads to variations in purity of the distillate, number of theoretical plates, reflux ratio and energy consumption which consequently leads to variation in cost. The VLE data predicted by the existing thermodynamic models show deviations from the experimental data. Hence, an Artificial Neural Network (ANN) model has been developed to predict the VLE so as to minimize the deviations from the experimental values. Several binary systems (1 simple and 6 azeotropicsystems) have been considered and VLE data has been predicted using the ANN, Margules and the van Laar models. The Root Squared Mean Deviation (RSMD) of predicted values has been calculated with respect to the experimental values. It has been observed that the data predicted by the ANN model is more accurate as compared to the Margules and van Laar models.