Novel modeling and optimization framework for Navy Blue adsorption onto eco-friendly magnetic geopolymer composite

Abstract

The disproportionate potency of dyes in textile wastewater is a global concern that needs to be contended. The present study comprehensively investigates the adsorption of Navy-Blue dye (NB) onto bentonite clay based geopolymer/Fe₃O₄ nanocomposite (GFC) using novel statistical and machine learning frameworks in the following steps; (1) synthesis and characterization of GFC, (2) experimental testing and modelling of NB adsorption onto GFC following Box-Behnken design and three response surface prediction models namely stepwise regression analysis (SRA), Support vector regression (SVR) and Kriging (KR), (3) parametric, sensitivity, thermodynamic and kinetic analysis of pH, GFC dose and contact time on adsorption performance, and (4) finding global parametric solution of the process using Latin Hypercube, Sobol and Taguchi orthogonal array sampling and combining SRA-SVR-KR predictions with novel hybrid simulated annealing (SA)-desirability function (DF) approach. Under the given testing range, parametric/sensitivity analysis revealed the critical role of pH over others accounting ∼37% relative effect and primarily derived the NB adsorption. The statistical evaluation of models revealed that all models could be utilized for elucidating and predicting the NB removal using GFC, however, SVR accuracy was better among others for this particular work, as the overall computed root mean squared error was only 0.55 while the error frequency counts remained <1 for 90% predictions. GFC showed 86.29% NB removal for the given experimental matrix which can be elevated to 96.25% under optimum conditions. The NB adsorption was found to be physical, spontaneous, favorable and obeyed pseudo-2nd order kinetics. The results demonstrate the suitability of GFC as the promising cost-effective and efficient alternative for the decolourization of urban and drinking water streams and elucidate the potential of machine learning models for accurate prediction & elevation of adsorption processes with less experimentation in water purification applications.

Keywords: Desirability function; Fe(3)O(4) nanocomposite; Kriging; Modelling; RSM; Simulated annealing; Stepwise regression; Support vector.