Hybrid Modeling and Predictive Control for an Agglomeration Process

Process simulation is an increasingly useful tool for the analysis of processes in the mineral industry. The ability to simulate process behavior without having to disturb plant operation prevents the loss of man-hours and production. Additionally, simulation tools provide a platform for the development of control and optimization strategies. Model predictive control, in particular, relies heavily on precise models to make accurate predictions. In mineral processing, due to complexities such as strong nonlinearities, variable coupling, time varying parameters, etc., the development of accurate process models becomes an increasingly difficult task. Furthermore, current control solutions mostly use linearized models and rely on expert systems to handle discrete variables such as discrete valves, signaling lights and tripper cars. Controllers that do not take into account the operation of these discrete variables may exhibit a suboptimal performance. For the previous reason, it is of special interest the design and development of hybrid controllers, capable of incorporating both continuous and discrete variables. This paper firstly describes the modeling and simulation of the main processes of a hydrometallurgical plant, followed by the hybrid model identification of the agglomeration process. Simulation results of the identified models are compared to real plant data to qualitatively validate the models developed. The hybrid models identified in this manner are used for the design and development of hybrid controllers for the agglomeration process. Simulation results show that the hybrid controllers exhibit a better performance when compared to an expert control scheme.