Application of Machine Learning Algorithms in the Optimization and Classification of Crushing, Grinding, and Mineral Separation Circuits
Application of Machine Learning Algorithms in the Optimization and Classification of Crushing, Grinding, and Mineral Separation Circuits
sina soleimani1 mahmod eskandari nasab2
1) M.Sc. Student in Mineral Processing Engineering, Shahid Bahonar University of Kerman
2) Ph.D. in Mineral Processing Engineering, Shahid Bahonar University of Kerman
IOI Code
:
ABGG-ABKHKE
Publication :
11th.International Conference on Researches in Science & Engineering & 8th.International Congress on Civil, Architecture and Urbanism in Asia(11icrsie.ir)
Abstract :
Ore crushing and separation, as the most energy-intensive and costly stages of mineral processing, have consistently been a major focus of optimization research. In recent years, the application of machine learning methods for surrogate modeling, online classification, and multi-objective optimization of these circuits has rapidly expanded.
This paper presents an adaptive review of four recently published studies focusing on Semi-Autogenous Grinding (SAG) mills, double-roll crusher circuits, X-ray fluorescence (XRF) sensor-based sorting, and the identification of operating modes in crushing circuits. Their quantitative results are presented in the form of tables and graphs.
The findings indicate that gradient-boosting ensemble tree models, particularly CatBoost, and Extremely Randomized Trees (Extra Trees) demonstrate superior performance in predicting continuous process variables compared with linear regression methods and simple neural networks. Among the four multi-objective optimization algorithms investigated—NSGA-II, MOEA/D, RVEA, and SPEA2—the evolutionary NSGA-II algorithm achieved the highest Hypervolume.
Furthermore, the combination of Support Vector Machine (SVM) with Particle Swarm Optimization (PSO-SVM) achieved an accuracy of 93% in XRF-based sorting of copper ore, along with a notable improvement in Net Smelter Return (NSR) compared with linear regression methods.
Finally, a conceptual framework is proposed for integrating these approaches into a general **“Data → Preprocessing → Machine Learning Model → Optimization/Decision”** pipeline. The existing limitations and potential directions for future research are also discussed.
Keywords : crushing multi objective optimization machine learning soft sensor support vector machine (svm) random forest semi autogenous grinding (sag) mill sensor based sorting