Biooxidation of refractory sulfide-bearing ore using feroplasma acidophilum: Efficiency assessment and machine learning based prediction

Abstract

The adhesive properties of microorganisms on the surface of minerals play an important role in the biooxidation efficiency of sulfidic refractory gold ores. In this research, the simultaneous effects of monosaccharides, ore content, pyrite content, and time on the activity and growth rate of Ferroplasma acidiphilum-from native Acid Mine Drainage (AMD)- was investigated during biooxidization alongside finding the best machine learning approach for the prediction of process efficiency using the independent variables. The results revealed that the optimum condition for reaching the highest pyrite dissolution (∼75 %) is 15 days of operating time, pyrite content of 7.2 wt%, and ore content of 5 wt%, pH of 1.47, and D-+-sucrose, D-+-galactose, and D-+-fructose concentrations of 0.52, 0.09, and 0.12 wt%, respectively. The results of the model comparison indicated that the Artificial Neural Network (ANN) model was able to predict the experimental results of this study with acceptable accuracy and better than Genetic Programming (GP) and Polynomial Regression informed by Response Surface Methodology (PR-RSM) from experimental data. Finally, the results showed that the change in D-+-fructose and D-+-galactose concentration has no significant effect on ferric ions concentration and pyrite dissolution content, while the influence of alteration in D-+-sucrose concentration is significantly high.

Keywords

• Gold recovery
• Biooxidation
• Experimental data
• Mchine learning
• Genetic programming
• Artificial neural network model