Artificial intelligence and machine learning models for predicting the metallurgical performance of complex sulfide ore flotation process

This research study proposed a novel approach utilising AI models to predict the metallurgical performance of complex sulfide ore flotation. Five machine learning and artificial intelligence models were employed in this study, that is, Random Forest (RF), Artificial Neural Networks (ANN), Adaptive Neuro Fuzzy Interference System (ANFIS), Mamdani Fuzzy Logic (MFL) and Hybrid Neural Fuzzy Interference System (HyFIS). Sixty-two flotation tests were conducted on samples containing galena, chalcopyrite and sphalerite as the main valuable minerals, and pyrite as the main gangue mineral. Different variables were used as inputs in the AI studies including physiochemical and operational parameters. The flotation recovery of lead and copper and their corresponding grades in the bulk concentrate were the primary dependent variables (outputs). The input variables included the dosages of sodium cyanide (pyrite's depressant), sodium isopropyl xanthate (collector), zinc sulfate (sphalerite's depressant) and Methyl isobutyl carbinol (MIBC, frother); air flow rate; flotation time; and the speed of the impeller in the flotation cell, which is indicative of the energy input. For the purpose of AI model development, datasets were divided into two subsets. The first subset was primarily used for the training phase, and it comprised 80% of the total data. The second subset, consisting of 20% of the total data, was used for testing. The models’ performance was assessed using two main indicators: R-squared (R 2 ) for the proportion of explained variation and RMSE for the average prediction error. The Hybrid Neural Fuzzy Interference System demonstrated superior performance in predicting the recovery and grade of copper and lead, with R² and RMSE of 0.9895 and 1.069 for the training phase, respectively, whereas for the testing step the respective values were 0.9128 and 2.859.