Modelling of continuous centrifugal gravity concentrators using a hybrid optimization approach based on gold metallurgical data

Sakuhuni G., Emre Altun N. E., Klein B.

Minerals Engineering, vol.179, 2022 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 179
  • Publication Date: 2022
  • Doi Number: 10.1016/j.mineng.2022.107425
  • Journal Name: Minerals Engineering
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, CAB Abstracts, Chemical Abstracts Core, Communication Abstracts, Compendex, INSPEC, Metadex, Veterinary Science Database, Civil Engineering Abstracts
  • Keywords: Continuous centrifugal gravity concentration, Artificial neural network, Genetic algorithm, Modelling, Optimization
  • Middle East Technical University Affiliated: Yes


© 2022 Elsevier LtdConventional optimization procedures for Continuous Centrifugal Gravity Concentrators (CCGCs) impose a priori weighting on objectives and results in a single optimum point, thereby failing to exploit the competing grade/recovery relationship. To generate an optimum operating line rather than a conventional optimization approach, a non-dominated recovery/grade line has been developed using a novel multivariate optimization technique code. The goal is to simultaneously maximize both grade and recovery objectives. This optimization has been developed and tested using a Knelson Continuous Variable Discharge concentrator, CVD on flotation tailings of a metallic sulphide ore (0.7 g/t Au), without a priori weighting of the competing objectives. This paper outlines the developed technique, which is an iterative hybrid performance improvement approach that exploits Pareto's non-dominated approach to multi-objective optimization, for determining the operating curve for CCGCs. This work is entirely and directly based on gold metallurgical data (i.e. Au recovery and grades) rather than an indirect proxy to generate the optimum operating line for the Knelson CVD. As a part of the suggested approach, the Artificial Neural Network (ANN) simulation has been effective in predicting the CVD performance: for the experimental and ANN simulated results, correlation coefficients are 0.92 and 0.81 for the Au upgrade ratio and Au recovery, respectively. The prediction level of the Regression Model was acceptable for the Au recovery (R2 = 0.78) and good for the Au upgrade ratio (R2 = 0.88). Based on these, the developed Pareto solution and the hybrid optimization model also enabled significant prediction of CCGC performance: at different operational conditions Au recoveries predicted by the Pareto optimum solution (11–32%) correlated well with experimental results (9–26%). Further, the method enables assessment of the level of influence of operational parameters on the metallurgical outcomes, hence identification of an optimum grade vs. recovery operating line. Overall, the results demonstrate the robustness of the approach in optimization of CCGCs operation despite trace occurrence of Au. Suggested approach is highly reliable for tuning multi operation variables of CCGCs to enable operation of these unique machines to yield successful concentration performance in plant scale.