ABSTRACT
This study utilizes artificial intelligence and statistical modelling to optimize the operating parameters of a carbon-based electro-Fenton process for purifying model dye (RB19)-contaminated wastewater. Multilevel experimental Box-Behnken and uniform deisgns (BBD, UD) with four variables were analysed using polynomial regression analysis (PRA) and artificial neural networks (ANN), while the process optimisation was done using desirability function. For the given testing range but different design matrices and runs, both designs predicted a maximum RB19 removal (RB19-RR) of 90 ± 2.1% at lowest energy consumption (EC) of 0.44 ± 2.5 Wh, when voltage, Na2SO4, FeSO4, and time were maintained as follows: 4-5.3 V, 7-11 mM, 0.4-0.6 mM, and 35-40 min, respectively. All the design-model combinations portrayed the similar senitivity analyses, revealing that RB19 degradation and EC are primarily influenced by electrolysis time and voltage. The performance assessment demonstrated that all the design-model combinations also excellently predicted for unseen conditions as the maximum root mean squared error (RMSE) value for RB19-RR was 4.07, while it was 0.072 for EC, however, BBD-ANN performance proved to be slightly better than others. Having â¼57% less experimentation, UD based models managed to accurately predict the results for unseen conditions as the statistical errors were quite insignificant, even in some cases, RMSE found to be less for UD compared to BBD, elucidating the potential of uniform design as an alternative of conventional factorial designs. Nevertheless, the prediction accuracy is also dependent on modelling approach, as in some cases ANN failed to predict the response precisely specially when dealing with small data. Furthermore, techno-economic evaluation results spell out the efficacy of carbon felt based enhanced electro-Fenton process as promising environmental remediation technology and highlight its practical implication from view of operational cost.
