Multi-Objective Optimization of a Mineral Processing Plant via Machine Learning and Genetic Algorithms

"Recovery and grade of the ore produced are two important metrics, which define the profitability of a mineral processing plant. Achieving high grade and high recovery at the same time in a mineral processing plant is difficult due to conflicting nature of the two objectives. To explore the operability limits of the plant operations and find the set of operating conditions that can help achieve the best possible recovery and grade, there is a need to solve a multi-objective optimization (MOO) problem. Past research focused on building a plant-wide simulation model or on optimization of individual units or a part of a mineral processing plant like grinding circuit, crushing circuit, etc. However, it is necessary to consider optimization of the complete plant in order to optimize the plant performance. Here, we have formulated the mineral processing plant performance optimization as a MOO problem. We have applied machine-learning algorithms such as random forest to build models to predict the recovery and grade of the ore. A non-dominated sorting-based multi-objective evolutionary algorithm called NSGA-II (Non-dominated Sorting Genetic Algorithm II) was employed to solve the MOO problem. Since the number of parameters that influence recovery and grade are quite large, we have applied machine-learning algorithms for feature selection or identification of key process variables. Pareto optimal conditions were determined for manipulated variables like grinding mill throughput, speed ratio at SAG mill, speed of centrifugal pump, water addition at different locations etc. The results obtained are useful in identifying the operability of a mineral processing plant to achieve the optimum grade and recovery for a given feed grade and the processing circuit. INTRODUCTION Copper is majorly mined or extracted as copper sulfides from large open pit mines in porphyry copper deposits. Among all the available ores, sulfides are the most commercial ores, especially chalcopyrite (Greenwood and Alan, 1997). Copper grade in these ores generally lies between 0.4 to 1.0%. In order to perform smelting operation, the grade of copper has to be improved to around 30% (Biswas et al. 2013). This targeted copper grade is achieved through two stages of mineral processing. Firstly, the copper in the ore has to be liberated by crushing the mineral rock. Secondly, the liberated copper in the crushed ore is then physically separated from other minerals like nickel by froth flotation techniques."