Artificial neural network based modeling and simulation of spiral wound Nano-filtration module and analyzing input responses for removal of Arsenic (Ⅴ) from potable water

Abstract

An increase in heavy metal contaminants such as arsenic (V) in potable water is a major health hazard and is a cause of fatal diseases and several body disorders. A reliable remedy to tackle this problem is Nano-filtration membrane, which is economical and does not allow heavy metal ions to permeate. However, input parameters should be set to an accurate value for better results, which is quite difficult. This study addresses the complexity of employing Artificial Neural Network (ANN) to model the Permeate Flux and percentage Rejection of a Nano-filtration membrane using the Deep Learning Toolbox in MATLAB. Initially, the number of neurons in the hidden layer were optimized and deployed for better results. The minimum value of the MSE (0.001325) was achieved with 10 neurons in the hidden layer. The developed model provides a coefficient of correlation (R) of 0.98022, which signifies a good-trained model. The trained ANN was then simulated to verify the model after the validation, effect of various input responses on Permeate Flux and percentage Rejection were studied. The best working range for T, TMP, Conct_F and pH would be 32 °C–23 °C, 7.5 bars–2.5 bars, 0.8 mg/l to 0.5 mg/l and 8 to 4.5, respectively when Flux would be kept under 40 l/m²h and rejection would be above 75 %.