The identification of perturbations in a base metal flotation plant using computer vision of the froth surface
The appearance of the surface froth in the zinc roughers of an Australian base metal flotation plant was quantified on-line using textural image analysis. The textural features of the froth were characterised using the neighbouring grey level dependence matrix method. Poor correlations between the image features and the zinc grade over the medium to long term were observed, which were attributed to stochastic changes in the ore mineralogy, the long time lag between the points of image capturing and sampling by the on-stream analyser, and vibration experienced by the camera support structure. However, significant correlations were obtained when short term perturbations in the zinc grade and reagent addition rates were compared with changes in the image features, as summarised by an SOM neural net. For the complex base metal plant it was not possible to correlate the absolute value of the image features with the zinc grade or any other plant parameter. It was observed that the direction of change in the froth appearance, quantified by changes in the image features, was of significance. This implies that operators should be observant of perturbations in froth structure rather than controlling the circuit to obtain a specific froth type. A clear relationship between the perturbations in the grinding circuit and changes in the image features was observed, which implies that image analysis could be used as a diagnostic tool. © 1997 Elsevier Science Ltd.