EFFECTIVE USE OF BUBBLE SIZE DISTRIBUTION MEASUREMENTS 1
G. R. Forbes1, G. de Jager2 and D.J. Bradshaw1 Mineral Processing Research Unit, University of Cape Town, Cape Town, South Africa 2 Digital Image Processing Group, University of Cape Town, Cape Town, South Africa
[email protected]
Recent years, lots of interesting the froth phase, machine vision technology was developed to study it, because it is not intrusive and real time. Many plants use it for online control, but there are still problems. Recent developments in machine vision systems enable the accurate measurement of bubble size distributions for a wide range of froths, including copper, platinum, coal, etc. With the increased availability of computational power, one can now effectively process images of froth which contain both large and tiny bubbles simultaneously. These were traditionally difficult to segment and resulted in incorrect bubble size distributions. For certain froths, very large differences exist between bubble size distributions of two images taken a fraction of a second apart (see Figure 1). This clearly indicates that an analysis of a single image of froth to determine a bubble size distribution is insufficient to provide an accurate indication of the state of the froth phase. The timing and frequencies of these measurements are crucial. Despite this, all of the currently available image analysis techniques still rely on the analysis of a single frame in order to produce a bubble size distribution.
A
B
Figure 1. Two images of the same froth taken approximately 1 second apart (under steady state conditions) and their cumulative bubble size distributions. This work evaluates a number of existing methods for comparing the similarity of distributions and determines which methods are better suited for this task. A suitable distance metric is obtained for bubble size distributions, which enables both supervised and unsupervised classification of the bubble size distributions. These classified distributions are linked to metallurgical sample data that was obtained from an industrial flotation cell, with sampling matching the time the bubble size distribution was determined.