Treatment of phenolic effluents by a thermochemical oxidation process (DiCTT) and modelling by artificial neural networks

Abstract

A Direct Contact Thermal Treatment (DiCTT) process was developed in this study, whereby the optimal conditions were identified for the thermochemical oxidation of phenol in a pilot plant. The following operational parameters were considered: the molar stoichiometric ratio of phenol/hydrogen peroxide (R) at a constant feed flow rate of the liquid effluent constant of (QL)=170Lh−1; the power dissipated by the burner (P) of 38.6kW at 10% excess air (E); an initial phenol concentration (Cph0) of 500mgL−1 and a combustion gas recycling rate (VR) of 50%. The thermochemical oxidation of phenol was monitored from oxidative degradation to the mineralization of the organic compound and the formation of acids; phenolic intermediates formed during the oxidative process were analysed. The concentrations of phenol, catechol, hydroquinone and para-benzoquinone were monitored by High Performance Liquid Chromatography (HPLC), Total Organic Carbon (TOC) was measured with a TOC analyser and the hydrogen potential (pH) was measured using a pH meter. Artificial Neural Networks (ANNs) were also used to model the experimental degradation results of the organic pollutant from the DiCTT process. For the ANN modelling, “Statistic 8.0” software was used with a “Neural Networks” module to predict the phenol concentration as a function of time. An R value of 75% was identified as the optimal condition for complete phenol degradation and a TOC conversion of approximately 60%; these rates are satisfactory for phenol degradation, although mineralization rates could be improved. The ANN modelling predictions for the phenol concentration as a function of time showed that the data were best described by a regression model in the form of a Multi-Layer Perceptron (MLP) network. The MLP consists of three layers with two neurons in the input layer, 7 neurons in the hidden layer and three neurons in the output layer. The correlation coefficients (R2) between the network predictions and the experimental results were greater than 0.99, indicating that the model was satisfactory.