Implementation of a Multivariable Controller for Grinding-Classification Process

The operation and control of grinding-classification circuits is usually based on single PID loops, and although it is recognized as a multivariable process with strong interactions, mainly due to their slurry recirculation, it is a fairly stable process from the operational point of view. This paper describes the implementation of Robust Multivariable Predictive Control Technology (RMPCT) in Codelco Chile El Teniente Division, leading ultimately to the control of a grinding-classification circuit, using the sump that separates the ball mill and the classification stage as a transients disturbance filter as well as finding an optimal operating point based on an objective function for maximizing throughput with a specified product size. Additionally, the implementation is done based on closed loop process identification, given an initial “seed model” that was duplicated on similar grinding circuits, reducing the controller implementation time.