Improved Control of Phosphate Flotation by means of a Simple Mineralogical Regression Model

A simple mineralogical regression model is described, which was devised in order to help the flotation plant cope with rapidly changing headgrades and feed mineralogy. Minerals in the feed were determined by means of quantitative X-ray diffraction and compared with plant performance parameters and flotation reagent consumption rates, using regression analysis. A mineralogical regression model for predicting recovery served to motivate flotation plant personnel and also set standards whereby flotation performance could be measured. The reagent consumption model indicated in which direction any or all of five flotation consumption rates should be altered. All practically measurable parameters, even those showing poor or no correlation, were included in the model, provision being made for the continuous update of a batch of data of 30 lines. Compared with the method of multiple linear regressions, the r2 is poorer at 0,41 (versus 0,50), but the mean standard error of estimate for both methods is identical at ± 6,6 units. The regression model is simpler than the method of multiple linear regressions, and although the accuracy of prediction is not high (41%), the mineralogical regression model proved to be successful in practice. Flotation efficiency, as measured by the mean ratio of head to tails grade, for the 9 month period after the mineralogical regression model was implemented on an hourly basis, improved from 2,8 to 3,2, which is equivalent to an improvement in average flotation recovery of 5% relative.