Hybrid ANN-PSO modeling and risk assessment approach of molybdenum separation from tungsten in sulfate solutions by liquid–liquid extraction process with D2EHPA

Abstract

The liquid–liquid extraction process for Mo(VI) in the presence of W(VI) from a sulfate solution with di-(2-ethylhexyl) phosphoric acid (D2EHPA) was studied using an artificial neural network (ANN) model coupled with a particle swarm optimization (PSO) algorithm. This study examined the influence of several operating parameters, including equilibrium pH, mixing time, reaction temperature, extractant concentration, and organic/aqueous (O/A) ratio, on the selective extraction of molybdenum from tungsten. In addition, the effects of different concentrations of ammonium bicarbonate on the metal stripping process from the loaded organic phase were analyzed. Subsequently, intelligent predictions regarding the molybdenum separation from tungsten were made using the ANN methodology in conjunction with the PSO technique. The numerical modeling results indicated that the coupled ANN-PSO approach demonstrated superior predictive performance, as evidenced by a higher correlation coefficient (R2) and lower absolute average relative error (AARE), in comparison to the simple ANN approach for predicting the selective extraction of molybdenum from tungsten. A Monte Carlo simulation (MCS) using a pseudo-random algorithm was employed to evaluate the uncertainty principle associated with the measurement-based input parameters that affect the separation factor response. The PERT distribution function was used to conduct the sensitivity analyses by incorporating mathematical formulae to generate sequences of random numbers. The findings of these analyses revealed that the factors with the greatest influence, arranged in descending order, were the equilibrium pH, D2EHPA concentration, O/A ratio, mixing time, and reaction temperature.