Modeling the relationship between forward osmosis process parameters and permeate flux

Abstract

Artificial neural networks (ANN) models are becoming more popular than mathematical and transport-based models due to their high performance and accuracy. Previous literature shows a lack of application of powerful ANN techniques for predicting forward osmosis (FO) performance. In this study, we developed a feedforward network to predict and analyze the permeate flux in the FO process. The ANN model was developed based on a lab-scale experimental database from various published articles. The permeate flux was modeled as a function of membrane-type, membrane orientation, feed and draw solution molarity, feed and draw velocity, molecular weight, feed solution temperature, and draw solution temperature. The influence of foulants on permeate flux has not been considered for developing the ANN model to avoid over-complication of the present work. The adj. R-squared values for train, unseen test and total datasets are 0.99, 0.92, and 0.95, respectively. These values are higher than those found in previously published literature (0.97, 0.85, and 0.82). Moreover, this is the first time that the effect of individual variables on permeate flux has been estimated quantitatively.