(100g) Experimental Validation of Lattice-Boltzmann CFD Using 2-D PIV and Torque Measurements for Fully Baffled and Unbaffled Systems

Pharmaceutical companies often deal with high complexity mixing processes involving crystallization and multi-phase systems. Often, vessel shapes or impeller configurations for which insufficient pre-existing literature is available are employed due to the specialist requirements of the pharmaceutical industry. Rather than require that new exploratory research be performed prior to employing these set-ups a broader solution is recommended. However, these factors present a problem for commonly used CFD software due to inherent complexities of multi-phase systems and high computational overheads. M-Star is a newer software package that combines a Lattice-Boltzmann-based fluid transport kernel with a high-resolution flux limiter scheme to solve the Navier-Stokes and advection-diffusion equations simultaneously. This work aims to further prove the viability of this approach in comparison to standard CFD packages based on discrete element approaches which are more computationally “expensive”. The overall aim is to create predictive models for non-standard vessel configurations for power number predictions, free surface movement detection and multi-phase mixing.

Initially in order to validate the use of M-Star CFD software, a simulated run was created reproducing conditions and CAD models from readily available literature, standard cylindrical vessels mixing with a Rushton impeller, using ANSYS Fluent. This initial run compared the outputs of the two software, once steady state was achieved. Further, using a lab torque meter, torque values and corresponding power numbers were recorded at several different stirrer speeds as well as various H/T values (1.0, 0.75, 0.5) to test M-Star time accurate and surface simulation capabilities. We will present the results comparing the values from the different approaches and in doing so, validating the software. Further experiments were then carried out, focusing on geometries and setting recreated within the lab. A singular flat bottom and open top cylindrical tank was used. Torque measurements were acquired when running a standard 6-bladed Rushton impeller of 0.1 m diameter at 50 RPM intervals up to 300 RPM. This experiment was reproduced in both a fully baffled (4 baffles) and un-baffled system. Lastly for every set-up, 2-Dimensional Particle Image Velocimetry (PIV) was used to capture mean velocity magnitudes and vorticity within the vessel.
Results observed compared experimental values acquired using the torque meter as well as general trends in velocity magnitudes within the vessel with those acquire from the constructed simulations. The PIV results were also used prior to each individual simulation run in order to set-up model probe locations at points of notably high vorticity. Further research to be carried will focus on reproducing more complex multi-phase systems followed by non-standard configurations often found in industry. Of interest is the accuracy of the simulated surface behaviour as well as transitional flow regimes at lower Reynolds numbers, where it is understood LES simulation methods lack accuracy. The aim will be to prove the validity of lower resolution CFD predictions even for multiphase flows in non-ideal geometries and thus to establish its applicability to such complex systems. This work is aimed at aiding pharmaceutical process scale-up of lab tests to large batch production.