Identification of the Dynamic Behaviour of an Autogenous Mill by use of Time Series Cluster Analysis

"The use of autogenous mills has become widespread, owing to their capacity and favorable capital and operating costs. Optimal operation of these systems is critical, given their high energy consumption and effect on downstream processing. A considerable effort has been spent over the years in the development of advanced control systems to achieve a better performance in this regard. However, despite significant advances, the problem still remains challenging, owing to the nonlinear behaviour of the system with respect to the mill variables, and, the unmeasured ore variability come from the mine.The mill load, in particular, can be seen as a state variable of the mill, which could be important for; modelling and control, and, important insights into the mill behaviour. In this investigation, time series cluster analysis was performed using online mill load measurements in order to identify the different control states of the mill. More specifically, this time series was divided into short non-overlapping segments. A distance matrix was set up for each segment and the information was captured by means of textural features, similar to that used in multivariate image analysis. These wavelet and grey level cooccurrence matrix features could subsequently be used to rapidly identify the normal operating conditions from the feed disturbances in the mill.In principle, the methods and models presented here, once suitably validated and calibrated, could be used in online process monitoring of autogenous mills. They also could serve as a foundation for developing more advanced process control models."