On-Line Identification and Classification of Grinding Mill Behaviour and Optimising Trajectories

"To remain competitive in an open market, mining companies are more and more requested to optimise their complex and highly multivariate processes. The increasing amount of process data, brought about by intensive use of instrumentation and the progress in computer data acquisition, enhances the need for more and more powerful and robust tools for the analysis and modelling of the process data. Now, with the emergence of multivariate data analysis (MDA) techniques, on-line clustering of mineral process behaviour is possible and could help in the extraction of valuable process operation indicators. MDA can be used in different mineral processes such as flotation, grinding, smelting etc. In this paper, the focus will be on grinding process as it is the one of the most costly node in the mineral processing scheme (i.e. power consuming) and it is a highly multivariable process. Particularly for primary grinding mills (SAG/AG), more instrumentation outputs and more powerful and significant process indicators are needed for efficient control and optimisation.MDA is a novel approach suitable for understanding mineral process behaviour compared to the traditionally approaches based on empirical and phenomenological modelling. Multivariate techniques can be used for several purposes. In this paper, the following objectives will be addressed:operation diagnosis: Shift differentiation and operation test differentiation can be showed by using MDA tools such as principle component analysis and factorial discriminant analysis.operation performance monitoring: operation regimes can be evaluated by clustering based on an economic criterion performance such as costs, tonnage, product quality, profit, etcIdentification of operation behaviour (monitoring) and selection of drift correction trajectories (control):MDA can be used to identify periods of unusual or abnormal process behaviour and to diagnose, if it is possible, causes for such behavior."