Computational fluid dynamic modelling of stirred reactors : power, baffle stresses, mixing times and semi-batch precipitation
A commercial CFD (Computational Fluid Dynamics) code CFX (version 4.2 to 4.4) from
AEA Technologyl'' has been used to compute the fluid flow, power number, Po, the stresses
on baffles, mixing time and a precipitation reaction in a mixing vessel. The impellers
investigated were Rushton turbine and 4 or 6 blade 45° pitch blade turbine. The impeller
generated flow was modelled primarily using the sliding mesh technique, with additional
modelling using Multiple Frames of Reference (MFR) for the mixing time simulations.
The Po was estimated from three different methods i.e. specific energy dissipation rate, ET,
summation, torque acting on the impeller surfaces, POp(primary power number), and the
reaction torque acting on the vessel walls and baffles, POs (the secondary power number).
The Po from the summation of ET, was underpredicted as compared with experimental values
in all the simulations by over 50%. The investigation of the calculated power numbers for the
vessels found that the closest and most consistent values of Po compared to experimental
results were obtained from the torque acting on the impeller surfaces, POp. The value of POs
was found to be greatly dependent on the sliding mesh simulation parameters and an
improvement in the POsprediction could be obtained by using a small time step. A further
investigation lead to the computation of the tangential forces and subsequently the axial
pressure distribution on the baffles. The baffle pressure distribution depends on the impeller
type and its clearance and was better predicted for greater impeller clearances and for the
radial flow impellers.
The mixing times simulations were performed using a computational method analogous to the
experimental method of probe responses. The system was in the high transitional flow regime
(Re=8800) and a low Reynolds k-e turbulence model was used in the development of the flow
field. The simulations were compared with experimental results (based on decolorisation
technique) and to three different mixing time correlations giving mixing times at three
different levels of homogenisation (i.e. 90%, 95% and 99%). Worryingly, the simulation
results were found to depend on the radial feed position even though the experimental results
suggest that it does not. At certain radial position, the simulated mixing time responses
accurately predicted the mixing times from the experiments and empirical correlations. CFD
based flow visualisation showed that the feed position influenced where the majority of the
tracer was initially distributed. The further the radial position was from the axis of the
impeller, the more the bulk of the tracer moved towards the low velocity region near the
vessel walls, leading to an overestimate of the mixing time. The sliding mesh and MFR
simulations of the velocity fields were used for the computation of the mixing time. The
results were similar in each case.
The precipitation modelling was achieved through the coupling of the CFD hydrodynamics
and user defined precipitation model. This approach was able to predict the performance of a
semi-batch process involving the precipitation of BaS04 with 270 s addition time. The results
(i.e. mean crystal size (d[4,3]) and the particle size distributions) were compared with
experimental results for a double feed precipitation reaction for a number of feed
configurations and concentration ratios.
Overall reasonable trends and agreement have been obtained for the modelled Po, mixing
time and baffle stresses. The precipitation model was less successful and was very dependant
on the different crystal shape factors used in the simulation model. Further experimental
work is required in order to define this parameter accurately, especially as experiments have
shown that it varies during the addition time.