Visual information model based predictor for froth speed control in flotation process

Image processing sensors are emerging as an important measurement option in mineral processing, mainly due to their non-intrusive characteristics. Their principal application areas have been the determination of ore size distributions in grinding and froth features in flotation. The incorporation of visual information in control loops is the logical step. However, the excessive processing required brings a new problem that must be solve: to count with a strategy able to provide a measured value for each visual sensor in the plant. A first approach is to assign one computer to each sensor yielding a distributed architecture, but this means the implementation of a huge computer network. A more efficient alternative is alternated sampling, but the succeed of this option is limited to the existence of virtual sensors capable of give accurate values that must be use during the unsampled period. In this paper we begin by reviewing classical image processing algorithms used in flotation froth feature extraction. Then a new method is introduced for the characterization and recognition of visual information using dynamic texture techniques. Finally we developed a dynamic texture based virtual sensor for the prediction of froth speed in the unsampled period, tested with industrial data.