Prediction of forward osmosis membrane engineering factors using artificial intelligence approach

Currently, forward osmosis (FO) is widely studied for wastewater treatment and reuse. However, there are still challenges which need to be addressed for the application of the FO on a commercial scale. In the meantime, with a strong capability to solve the complicated nonlinear relationships and to examine of the relations between multiple variables, artificial intelligence (AI) technique could be a viable tool to improve FO system performance to make it more applicable. This study aims to develop an AI-based model for supporting early control and making decision in the FO membrane system. The results show that the artificial neural networks model is extremely suitable for prediction of water flux, membrane fouling, and removal efficiencies. The most appropriate input dataset for the model was proposed, in which organic matters, sodium ion, and calcium ion concentrations played a vital role in all predictions. The best model architecture was suggested with an optimal hidden layers (2–4 layers), and neurons (10–15 neurons). The developed models for membrane fouling show strong correlation between experimental and predicted data (with R² values for prediction of membrane fouling porosity, thickness, roughness, and density were 0.85, 0.97, 0.97, and 0.98, respectively). The prediction of water flux presented a high R² and low root mean square error (RMSE) of 0.92 and 0.9 L m⁻².h⁻¹, respectively. Prediction of the contaminant removal exhibits a relatively high correlation between the observed and predicted data with R² values of 0.87 and RMSE values of below 2.7%. The developed models are expected to create a breakthrough in the control and enhancement in a novel FO membrane process used for wastewater treatment by providing us with actionable insights to produce fit-for-future systems in the context of sustainable development.