Optimization of reagent mode selection of sulfide ore flotation

Introduction. This study is devoted to the application of machine learning methods to optimize the flotation process of gold ore. The objective of the work is to determine the optimal values of the controlled parameters used in the process to achieve the best target parameter, such as recovery. Materials and methods. The main optimized parameters in this work are the amounts of flotation reagents. The effect of different amounts of flotation reagents on performance indicators is not linear and has complex physical causes. For iterative optimization of the flotation process by monitoring its parameters, the authors turn to Bayesian optimization (BO) based on Gaussian process regression (GPR). The authors selected a set of adjustable parameters that can be controlled during the flotation process. They determined the concentrations of the reagents used as a set of such parameters. Based on expert knowledge, the boundary values of each parameter were selected and the initial vector of controlled parameters was determined to run the algorithm. A series of experiments and analytical studies of the quality of the obtained concentrates were carried out. The BO algorithm produced candidate parameter vectors based on the obtained objective function values at each iteration and attempted to maximize the objective function step by step. Results. TTo assess the convergence of the model, a graph was constructed comparing the predicted objective function values with the measured experimental results for several optimization iterations. The proposed method allowed for 10 algorithm iterations to bring the valuable component recovery in the concentrate to 89.53%. The underlying mathematical approach was described in detail by the authors, and the obtained results were confirmed experimentally. Discussion. The obtained data help to control the algorithm, but requires expert knowledge. The problem of parameter space exploration can be further simplified by constructing a surrogate model for recovery efficiency and following a multiple-validity optimization approach, which we leave for future work. Conclusion. The paper presents the results of studies on sulfide ore flotation with the determination of optimal values of the controlled parameters used in the process of obtaining the best value of the objective parameter using the Bayesian optimization method. Resume. The obtained results can be used in industry to increase the efficiency of gold ore processing and optimize production processes