 |
Liquid atomisation is an important operation in process industries, all kinds of liquid fuel combustion and several other contexts. The numerical simulation of such processes allows for their efficient optimisation as well as trouble-shooting.
The aim of such simulations is to predict the complex physical phenomena in spray atomization based on material properties and first principles where possible and affordable.
The disintegration of an initial liquid jet, sheet or film usually happens in multiple steps: First, the contiguous liquid breaks into ligaments, which then split further into multiple droplets. The droplets can, depending on the surrounding gas flow, undergo further, so-called secondary break-up.
Since the latter part only involves droplets in a surrounding gas flow, multiple semi-empirical (i.e. calibrated) models have been published for it that can be efficiently used in the Lagrangian formulation; this is a comparably “cheap” multiphase model approach that can be applied to relatively large flow simulation domains.
In contrast, the governing forces in the primary break-up are highly dependent on the individual atomizer design; a predictive simulation of this part of a spray is only possible using a “first principles” approach that resolves the shape and movement of the gas-liquid boundary in all detail. For that purpose, the Eulerian “Volume-of-Fluid” (VOF) multiphase simulation approach is recommended, which requires much higher mesh resolution and small time steps.
In order to achieve maximum modelling fidelity at minimum possible cost, the two approaches have been connected in the commercial general-purpose CFD code ANSYS Fluent by a model-transitioning mechanism. Its construction and some simulation results will be shown.
The use of dynamic solution-adaptive mesh refinement allows to accommodate the high-resolution mesh requirements for the VOF model at minimum possible cost. Still, small time steps need to be used for accurate interface tracking, which keeps the total computational cost considerable.
It is therefore suggested,in addition to the model-transitioning, to split the spray simulation into two parts.The first part covers in-nozzle flow and/or the near-nozzle dense spray region and does the model transitioning. From that, information about the Lagrangian particle parcels is exported and transferred into the second simulation part. This allows much larger simulation domains where the spray is considered dilute and secondary break-up can be modelled.
The use of this cost-saving split approach will be demonstrated, including special means to flexibly control the discretisation errors in the Lagrangian simulation of the dilute spray region.
