Coupling Dynamic Data Reconciliation with Model Predictive Control for Real-Time Optimization of a Flotation Plant Simulator

Maximizing benefits generated by separation plants in the mineral processing industry is of prime importance. To achieve this goal, observers and controllers are both key tools playing a crucial role for economic optimization. Although these methods have been widely discussed in the literature and that numerous applications have been reported, coupling the different algorithms raises practical challenges. The economic function definition, the selection of the real-time optimization structure, the elaboration of observer and controller models, the controller development, and the tuning of stochastic and deterministic parameters strongly influence economic performances. The objective of the paper is to discuss the major difficulties related to the implementation of a real-time optimization system and highlight the benefits of linking data reconciliation and model predictive control for a simulated flotation plant. A systematic topdown approach is used to go through the main design phases and to comment on the different options available for the development of the optimization structure. An internal model predictive controller coupled with a stationary observer is applied for the direct optimization of a three-stage flotation plant simulator based on a phenomenological model. The results show performance improvements in terms of increased revenues brought by reduced grade variability and collector consumption. The discussion also proposes alternative avenues that could be taken to achieve comparable improvements. Finally, since the limited simulated case-study cannot provide a complete understanding of the benefits brought by a data reconciliation observer, recommendations for industrial applications, which are more complex, are provided to promote effective implementations.