Adsorption of toluene from aqueous solutions onto polyethylene glycol modified bentonite: Kinetic, isotherm studies and artificial neural network modeling

Abstract

In this work, Algerian natural bentonite was used in the preparation of a new adsorbent by impregnation with polyethylene glycol (PEG) for the removal of toluene from an aqueous solution. These interactions have been monitored by thermal method (thermogravimetric analysis) as well as by Fourier-transform infrared spectroscopy. The adsorption efficiency of natural and PEG-bentonite was examined for toluene removal by batch adsorption experiments under different operating conditions. Therefore, a multilayer perceptron (MLP) neural network was then used to predict the adsorption capacity of toluene. Different training algorithms were compared to determine the most suitable training algorithm. A single hidden with six neurons using a tangent sigmoid function transfer with the Levenberg–Marquardt backpropagation algorithm has been found the best predictive performance. The high value of coefficient determination (0.999) and low value of root mean square error (0.00074) proved that the MLP model can predict the adsorption capacity of toluene with reasonable accuracy. Furthermore, the sensitivity analysis based on the MLP model indicated that the contact time and the initial concentration of adsorbate with the same relative importance a round of 39% appeared to be the most influential parameter in the adsorption capacity of toluene, followed by adsorbent dose (20.99%).