Spray Drying Technology.pdf - National University of Singapore

If the former case is encountered, it is then a good indication that the
less computationally expensive steady state solution is suitable for the model;
flow is inherently steady. However, although a converged solution might be
achieved in the steady state, it is important to check if the solution is
physically ‘logical’. Although not reported, it was found that certain
numerically converged solutions produce aberrant flow patterns. Such
aberrant behaviour was alleviated when the solution was changed to the
transient framework. On the other hand, the oscillating residual pattern as
mentioned above is definitely a good indication to switch to the transient
solution framework, using the semi-converged solution from the steady state.
Both of these numerical aspects were reported by Woo et al. (2009) and
Ullum (2006). In addition, one should also check if the flow field significantly
changes at subsequent iterations after ‘convergence’. If the flow field
changes significantly or even produces aberrant flow contours, then it is most
likely an indication of inherently transient behaviour.
Another key numerical aspect is in the refinement of the mesh used. It is
common CFD practice to perform mesh independence tests particularly for
regions of high gradients. This is normally undertaken by systematically and
progressively using smaller mesh sizes until the solution is independent of the
mesh size. However, for such possibly self sustained fluctuation flow, on top
of providing mesh independence to the solution, the mesh size also
determines whether or not the possibly transient flow structures are captured.
This was firstly reported by Oakley et al. [25] , where their steady axisymmetric
solution becomes progressively unsteady when the mesh size was reduced to
a certain high refinement. Although not specifically shown, Langrish et al. [5]
also observed such numerical behaviour in their Very Large Eddy Simulations
(VLES) of a pilot unit. Such a numerical behaviour is attributed to the ability to
capture smaller flow structures as too coarse a mesh will make the solution
too diffusive, thus damping out any transient behaviour. Woo et al. (2009)
further reported on this behaviour by using an extensive combination of mesh
sizes and chamber diameter [23] . In their three-dimensional simulation work on
a short-form co-current spray dryer, a total of 946071 mesh elements are
warranted for a transient solution whereas a mesh of only 288511 produces a
steady state flow solution without any self-sustained fluctuations.
Two dimensional versus three-dimensional
Consideration in using a two-dimensional (including axisymmetric models) or
three-dimensional simulation will greatly affect the computational time and
resources (Figure 4). Such requirements are further amplified in transient
simulations as the flow field needs to be developed and statistical averaging
might be required in the transient framework [21][26] . Most early work typically
utilizes the axisymmetric [25][27] and two-dimensional models [10-11] . In recent
reports, workers have shifted to three-dimensional simulations [23][26][28] . Apart
from the computational requirements, another important consideration
between the two approaches concerns the ability to capture any possible
transient behaviour in the flow. Guo et al. [20] have shown that transient
fluctuation, particularly of the central jet, is three-dimensional. To be more
precise, the jet tends to flap about a moving axis precessing around the
chamber. Further illustration of such fluctuation is illustrated by Woo et al.
using the jet feedback mechanism [23] . Using a two-dimensional model for
such flows will prohibit the precision of the model in capturing such important
Woo, Huang, Mujumdar, Daud – CFD modeling of spray dryers 7