The Use of a Laboratory Centrifugal Separator to Study Gravity Recovery in Industrial Circuits

"A laboratory Knelson Concentrator (LKC), 7.5 cm in diameter, was used to study the performance of a similar unit at plant scale, 76 cm in diameter, This paper reports on sampling and assaying procedures and problems, and discusses the performance of the plant unit.The sampling procedure was based on the extraction of large samples, 5-15 kg, and their complete processing with a LKC, to characterize size-by-size performance. The procedure yielded reliable estimates of overall gold assays, after blatant assaying errors were corrected. The plant centrifugal unit concentrate was sampled in situ, which yielded very erratic grades, with a large overall negative bias. Due to the very heterogeneous nature of the concentrate, which can be linked to flow patterns in the riffles of the unit, it is recommended that for future testwork, the concentrate be recovered separately, blended and then sampled.The plant centrifugal separator, when operated at a feed rate of 25 t/h, yielded total and free gold recoveries of 41 % and 62%, respectively. This is much lower than previously reported data, and suggest that at 25 t/h, the unit is overloaded. However, recovery did not decrease with time over a loading cycle of 2 hours, which suggests that present loading cycles, generally 1 hour, can be significantly increased. INTRODUCTIONThe study of gold gravity circuits is plagued with the statistical problem of determining the behaviour of particles occurring with a very low frequency. This problem becomes particularly acute with increasing particle size. For example, considering the average weight of a gold flake in the 840-1200 µm (20/28 mesh) to be about 5 mg (Banisi et al, 1991), a reasonably high grade feed of 10 git would require, for a relative error (standard deviation) of 10% on grade, 100 particles, that is 0.5 g of gold, or 50 kg of 840-1200 µm material (a Poisson distribution is assumed). Gy's formula, for the same data, yields a weight of 33 kg, with a shape factor of 0.1 and a size distribution factor of 0.5. Both results are comparable; to obtain the suggested weight of material, a stream containing 5% weight in the 840-1200 µm would require 1 tonne samples (of unsized material). The problem rapidly disappears with decreasing particle size. Still with Banisi et al's data, a gold flake in the 300-420 µm range averages 0.5 mg in weight; the same 10% error on grade corresponds to a weight of 5 kg. Gy's formula for the same size distribution and particle shape factors as above predicts a required mass of 1.4 kg. This would correspond to an unsized sample weight of 10 kg with 14% weight in the 300-420 µm, for example."