State and Ore Hardness Estimation in Semiautogenous Grinding ?

Semiautogenous milling is difficult to control both because of its non-linear behavior and the effects of overloading due to increases in the ore charge or variations in ore characteristics. Advanced control strategies and operational change detection methods are thus in need of strengthening using techniques such as state estimation. Non-linear state estimation is a complex task for which various solutions have been proposed, such as the extended Kalman filter, the particle filter and the moving horizon estimator. In this study we present firstly a quantitative comparison of these solutions using a dynamic model validated with mill data. The results indicate that in addition to its lower computational requirements, the extended Kalman filter delivers the best performance in robustness and estimation error. Next, we propose a method for estimation of changes in the hardness of the ore feed that we test by simulation. Finally, we show that this method also works with real data.