The behaviour of dense medium drum separators at Hamersley Iron Ltd's iron ore concentrator and Groote Eylandt Mining Company Pty Ltd's manganese concentrator was studied. Sampling and density tracer techniques were used to collect drum partitioning data over a range of operating conditions and medium compositions. Medium viscosity was measured during each partition test with a Debex on-line viscometer. Size and densimetric analyses of the drum feed and product ore samples were combined with the measured values of the medium properties and used to develop a new mechanistic model of drum partitioning behaviour.
A mass balancing method was developed in order to resolve inconsistencies in the size and densimetric analyses of the ore samples. The technique allows the adjustment of size, density and chemical assay data and calculation of the overall mass split in a single step. Non-linear fitting techniques were used to determine model parameters. A weighting strategy, based on theoretical sampling error, was developed in order to include knowledge of data accuracy in the mass balancing and model fitting procedures. A new published method for the estimation of density distributions using sparse densimetric data was evaluated and shown to have considerable potential for reducing densimetric analysis requirements.
The fitted partition parameters, Ecart probable (Ep) and separation density, were used to illustrate trends in drum partitioning behaviour. Ep was found to increase as medium density and medium viscosity increased, and to decrease as particle size increased. Separation density was found to be generally below that of the medium, a function of the bottom-fed separator, and to exhibit a minimum with respect to size at around 12mm. This minimum coincides with the size of the holes in the peripheral lifters of the drums.
Flow curves determined from Debex on-line viscometer data for a variety of dense medium samples indicated dilatant behaviour with the suggestion of a yield stress. A published procedure for the calculation of shear rate and shear stress from Debex viscometer data was found to be unsuitable for use at low shear rates. The flow behaviour of the dense media could therefore not be quantified at viscometer bobbin speeds below 300 RPM. Consequently, apparent equivalent Newtonian viscosity at selected bobbin speeds was used as the measure of dense medium viscosity.
A strong relationship between free settling particle terminal velocity and partition number was observed in the data. This relationship is almost independent of medium conditions, and is the basis of the new mechanistic drum model. The terminal velocity calculation combines medium density, medium viscosity, particle size, particle density and flow regime effects in a manner which is consistent with hydrodynamic laws. The terminal velocity model was shown to be capable of describing drum partitioning data as well as or better than a purely empirical approach. The new model has several advantages over empirical modelling techniques: it can be extrapolated confidently, it decouples the highly correlated effects of medium density and viscosity, and, because of its almost constant parameter values, it can be applied with relatively few data sets.
A simulation study of the Hamersley drum separator was carried out using the new model. It was found that a maximum in achievable product grade with respect to medium density occurs as a result of the opposing effects of medium density and viscosity on upgrading performance. The presence of magnetite or ore slimes in the dense medium in significant quantities therefore limits the ability of the drum separator to concentrate ore, particularly at higher medium densities.