Coal Preparation » Fine Coal
The aim of this project was to develop a generalised model of the Reflux Classifier (RC) utilizing a combination of a continuum approach based on the mechanisms of segregation and dispersion within the system, and a tracer particle approach based on the Discrete Element Method (DEM). The continuous segregation-dispersion model provides a direct solution to the transport of specific species through the system. Due to the computational demand of DEM, a tracer level approach was adopted in order to keep the total number of particles to a minimum, and thus make the computational demands manageable. Given the DEM operated at a tracer particle level, the degree of hindered settling was governed by the continuum solution, which was solved simultaneously with the DEM approach. DEM is based upon Newton's laws to describe particle translational and rotational motion, allowing the particle motion to be observed discretely. The segregation and dispersion approach, however, is based upon a volume flux balance method, and describes the net flux of a particle species relative to the vessel in terms of the dispersion and segregation contributions.
Validation of the model was undertaken with reference to the published literature. In 2008 it was discovered that the performance of the Reflux Classifier was improved significantly by introducing closely spaced inclined channels, with a nominal spacing of 6 mm (Galvin et al, 2010). The previous spacing had been typically 60 mm. This change led to the re-design of the technology, and the exploitation of a new laminar-shear mechanism, arising due to the high shear rate in the inclined channels. Therefore, following 2008, the variation in the D50 with particle size was reduced significantly and the Ep values also decreased significantly. For example, over the size range 0.25 to 2.0 mm the Ep was reduced from 0.15 to typically 0.08, well below anything achieved previously using water-based technology. The new model developed in this report focussed on this particular case, involving the laminar-shear mechanism, in the first instance, achieving excellent agreement with the reported partition curves (Galvin et al, 2010). The DEM model did provide some useful visualisation of the underlying mechanism, but the computational demands were considered too great for the model to be of general use.
Following the success of the model in describing the laminar-shear mechanism, and hence the partition curves in an RC2000 unit, additional validation work was conducted by simply removing the contributions of the shear rate in the inclined channels. Hydrodynamic dispersion is governed by the value of the dispersion coefficient which is applied in the vertical and horizontal directions of the vertical section of the Reflux Classifier, and in the tangent direction of the inclined channels. However, in the normal direction of the inclined channels the dispersion coefficient is quantified directly via the shear rate. Thus, the effects of the laminar-shear mechanism were removed by 'switching off' this contribution due to the shear rate. The net result was that the model predicted the poorer separation performance observed in the Reflux Classifier prior to 2008, with reference to published data in 2005 (Galvin et al, 2005). The agreement was excellent. Hence the model is considered to be both robust and predictive.
Overall, the continuum model based on the contributions of segregation and dispersion provides an effective model of the Reflux Classifier. The model describes the feed in terms of the mass portions within nominated size and density classes, for example six size fractions across seven density fractions corresponds to 42 species. Thus, the model is able to determine the partition curves for a given feed subjected to a specific volumetric and solids loading, and for a given yield, and fluidization rate. It is noted that beyond the size range of 0.25 to 2.0 mm the model showed some departure from the experimental data in the size range 0.125 to 0.250 mm. But as discussed in the report, the true behaviour may lie in between the experimental and simulated results hence this issue will be the subject of future study.