Modeling of fixed-bed dye adsorption using response surface methodology and artificial neural network

Abstract

In this study, artificial neural network (ANN) and response surface methodology (RSM) were applied to analyze the fixed bed adsorption of FD&C red 40 dye (Food, drug and cosmetic dye red 40) on polyurethane/chitosan foam (PU/CS foam). The adsorbent was prepared and characterized. The effects of the process variables, including flow rate and bed height, were investigated through two different levels using RSM. Breakthrough curves were used as training data set for the ANN. The ANN was customized with 10 neurons in the hidden layer using the hyperbolic tangent sigmoid transfer function as activation function and the linear transfer function in the output layer. The optimal range of bed operation was 5–6 cm for bed height and 15–17.05 mL min−1 for flow rate. The values of experimental adsorption capacity of the column ranged from 44.3 to 108.1 mg g−1, and were compared with the resulting values of the ANN and RSM models. The ANN can predict the experimental data with more accuracy than the RSM. The values found for the coefficient of determination ​​were 0.9911 for the ANN and 0.8853 for the RSM. The various error functions tested between predicted and experimental values ​​of the ANN and RSM models demonstrated a better applicability and efficiency of the ANN model. Finally, PU/CS foam proved to be a promising, low-cost adsorbent with excellent potential for removing FD&C red 40 dye.