A non-linear model-based control architecture for a single-stage grinding mill circuit closed with a hydrocyclone
is proposed. The control architecture aims to achieve independent control of circuit throughput
and product quality, and consists of a non-linear model predictive controller for grinding mill circuit
control, and a dynamic inversion controller to control the fast sump dynamics. A particle filter is used to
estimate the mill states, and an algebraic routine is used to estimate the sump states. The observers make
use of real-time continuous measurements commonly available on industrial plants. Simulation results
show that control objectives can be achieved by the controller despite the presence of measurement
noise and disturbances.