Machine Vision Measurements for Molybdenite Grade Modelling

Flotation froth machine vision systems provide consistent real-time measurements pertaining to the state of the flotation cell being monitored. Typical measurements include froth velocity, froth colour, froth texture and froth stability. This paper presents the results from test work conducted on the molybdenite rougher circuit at the Kennecott Utah Copper Concentrator during February 2006. Machine vision results show that it is not possible to accurately determine bubble size distribution information for the molybdenite froth, despite using state-of-the-art methods. Instead, it is necessary to make use of texture measurements to classify the various different froth states. Numerous combinations of texture measurements and classifiers are used to determine the best combination. Results show that the Fourier ring based texture measurements perform the best when both K nearest neighbour and Gaussian mixture model classifiers are used. The concentrate grade of the flotation cell is modelled using combinations of froth velocity and froth texture descriptors. The results show that by using a linear combination of these descriptors rather than either one by itself it is possible to account for a much larger percentage of the variation observed in the concentrate grade data. Adjusted R2 values of 92% and 84% are achieved when modelling the copper and the molybdenite concentrate grade respectively.