Multi-object optimization of Navy-blue anodic oxidation via response surface models assisted with statistical and machine learning techniques

This study aims to model, analyze, and compare the electrochemical removal of Navy-blue dye (NB, %) and subsequent energy consumption (EC, Wh) using the integrated response surface modelling and optimization approaches. The Box-Behnken experimental design was exercised using current density, electrolyte concentra-tion, pH and oxidation time as inputs, while NB removal and EC were recorded as responses for the imple-mentation and analysis of multiple linear regression, support vector regression and artificial neural network models. The dual-response optimization using genetic algorithm generated multi-Pareto solutions for maximized NB removal at minimum energy cost, which were further ranked by employing the desirability function approach. The optimal parametric solution having total desirability of 0.804 is found when pH, current density, Na2SO4 concentration and electrolysis time were 6.4, 11.89 mA cm(-2), 0.055 M and 21.5 min, respectively. At these conditions, NB degradation and EC were 83.23% and 3.64 Wh, respectively. Sensitivity analyses revealed the influential patterns of variables on simultaneous optimization of NB removal and EC to be current density fol-lowed by treatment time and finally supporting electrolyte concentration. Statistical metrics of modeling and validation confirmed the accuracy of artificial neural network model followed by support vector regression and multiple linear regression anlaysis. The results revealed that statistical and computational modeling is an effective approach for the optimization of process variables of an electrochemical degradation process.

Automated summary

This study analyzes and compares the electrochemical removal of Navy-blue dye and subsequent energy consumption using response surface modeling and optimization approaches. The optimal parametric solution for maximum dye removal and minimum energy cost is obtained through genetic algorithm optimization. Sensitivity analysis reveals the influential patterns of variables on simultaneous optimization, and statistical metrics confirm the accuracy of the artificial neural network model for prediction.