ISOTHERM MODELING OF MULTI-COMPONENT ADSORPTION OF BENZENE DERIVATIVES ONTO GRANULAR ACTIVATED CARBON

Abstract

The modeling study on simultaneous adsorption of multicomponent systems i.e. three binary systems (nitrobenzene-aniline (NI-AN), nitrobenzene-phenol (NI-PH) and phenol-aniline (PH-AN)) and one ternary system of nitrobenzene-aniline-phenol (NI-AN-PH) onto granular activated carbon (GAC) in aqueous solution were performed at 30 oC by conducting batch experiments. The single solute equilibrium adsorption data of nitrobenzene (NI), aniline (AN) and phenol (PH) were fitted with Langmuir, Freundlich and Redlich-Peterson model. The Redlich-Peterson and Freundlich model gave better fitting than Langmuir model for individual adsorption. The binary adsorption data were examined and compared by using adsorbed solution theory (AST) such as ideal adsorbed solution theory (IAST) model and real adsorbed solution theory (RAST) model. The ternary equilibrium adsorption data were predicted by using ideal adsorbed solution theory (IAST) model and real adsorbed solution theory (RAST) model and artificial neural network (ANN) models. IAST model doesn’t provide an acceptable prediction of binary data except for low liquid concentration levels, as it undervalued AN and PH adsorption capacity and overvalued NI in each binary system. Also for ternary system IAST model doesn’t provide an acceptable prediction of data except for NI, as it undervalued AN and PH adsorption capacity ternary system. This is due to the non-ideality of mixtures at high concentration levels in the solution. The RAST model gave an excellent prediction of binary adsorption experimental data. In case of ternary system RAST model gave better prediction than IAST. It was found that ANN model can predict the equilibrium adsorption data with minimum error using Levenberg-Maequardt back-propagation and Scaled conjugate gradient backpropagation algorithm out of six back-propagation algorithms and the transfer function used in hidden layer was either a log sigmoid transfer function (logsig) or tangent sigmoid transfer function (tansig) with 2 neurons and at output layer linear transfer function (purelin) was used. Since ANN provides excellent prediction thus it can be used as a reliable model for the design of industrial adsorption equipment.