Flotation of PGM-containing minerals: Plant-wide regression and prediction of circuit performance

• Plant-wide regression of flotation kinetics, including recycle streams.
• Excellent fit of model to plant data.
• Model capable of prediction of alternate circuit configuration performance.

A crucial step in the processing of PGM-bearing reefs in South Africa is the flotation of the sulfidic minerals from the host rock. Flotation plays a crucial role in reducing the mass of material to be processed further downstream by smelting, converting and hydrometallurgical refining. However, flotation has the lowest recovery of valuable metals along the processing chain. As a result, small improvements in efficiency in flotation have an enormous benefit to the company’s performance. Modelling and simulation play an important role in circuit optimisation, due to the ability to test and examine multiple options in a small amount of time without affecting the operational circuit. Unfortunately flotation is extremely difficult to model and simulate due to the complexity of the mechanism and the large number of variables involved. Whole circuit modelling is even more difficult, for the reasons already mentioned as well as maintaining a consistent mass balance in the face of multiple recycle streams. The purpose of this work is to construct a whole-circuit model of a flotation circuit, and use the model to predict the effect of changes to the flotation circuit. A goal in the construction of the model was to use as few parameters as possible. The work presented in this paper is focused on the regression of the parameters of a first-order kinetic flotation model to actual pilot plant data. The performance of the model was then validated by predicting the behaviour of a modified flotation circuit.The regression showed excellent correlation with actual plant data on a bank by bank basis. Using the regressed parameters directly on a modified circuit configuration, the model was able to predict the circuit behaviour when compared to plant data.