Performance optimisation of forward-osmosis membrane system using machine learning for the treatment of textile industry wastewater

Abstract Forward osmosis (FO) is an emerging low energy membrane-based technology and can be applied if the diluted draw solution (DS) is directly utilized without additional recovery/treatment. In this study, FO process was applied for the treatment of textile industry wastewater using fertilizer as DS. This paper focuses on modelling and optimisation of FO process using machine learning techniques like Response Surface Methodology (RSM), Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS). To model the FO process, central composite design was utilized to examine the effect of initial draw concentration, initial feed concentration, time, initial feed pH and temperature on the water flux and reverse salt flux. The optimum water flux (8.527 L.m−2.h−1 (LMH)) and reverse salt flux (7.246 g.m−2.h−1 (GMH)) was obtained using initial draw concentration of 1.625 M, initial feed concentration of 1090 mg/L, reaction time of 90 min, initial feed pH of 7.33 and temperature of 35 °C. Under these conditions, FO performance was carried out experimentally and validated with the models. The model developed for the FO process by ANN and RSM was considerably better than that of other models in terms of precision of predicting the water flux and the reverse salt flux, respectively. About six different chemical fertilizer solutions were screened and tested at optimum conditions to identify the best suitable fertilizer DS for FO process using textile wastewater. The results indicate that Diammonium Phosphate (DAP) along with Potassium Chloride (KCl) fertilizer as DS gave better performance with respect to water flux.