Lattice Boltzmann Methods on Parallel Supercomputers for Computing Mass and Momentum Transport of Foams in Filling Columns

点击翻译按钮获取中文摘要

Lattice Boltzmann methods (LBM) are established alternatives for fluid simulation, since they are well suited for efficient direct numerical simulation of complex scenarios. In the open source software waLBerla, many different lattice Boltzmann methods are implemented as well as their coupling to rigid body and multiphase models. The framework, designed in a systematic, performance-oriented development process, is used in a variety of different applications, such as material science, medicine and process engineering. waLBerla is also the basis of the mesoscale simulations that have already been carried out as part of a previous DFG/AiF cluster project `` Simulation von Proteinschäumen''.Naturally the scale transition from the single bubble to the full process results in enormous demand on compute power. To efficiently simulate these kind of scenarios, adaptively refined grids are used, that are available in the waLBerla framework already for single phase flows. In this project adaptive grid refinement and dynamic load balancing will be extended for free surface flows. Additionally, metaprogramming techniques will be used to write flexible, portable and maintainable code for modern hardware without performance tradeoffs.The resulting, highly efficient code will be used in this project to simulate foams in columns. Unwanted foaming in columns can lead to high pressure drops, a reduction in possible throughputs and reduced separation efficiency. To avoid these undesirable effects, a detailed understanding of the relevant foam creation and foam destruction effects is necessary. The numerical simulation can make a significant contribution here. On the basis of validated simulations of individual column fillers, the foaming behavior in a column is simulated and predicted. The main goals of this project are (i) increasing the performance of Free Surface Lattice Boltzmann (FSLBM) simulations through code generation techniques, adaptive grid refinement and dynamic load balancing, (ii) in close cooperation with the cluster partners, a virtual but experimentally validated design of processes and (iii) coupling waLBerla fluid simulations to CT reconstruction algorithms to improve geometry reconstruction.