Development and Validation of turbulence models for Lattice Boltzmann schemes: Fid-Bau Portal

Development and Validation of turbulence models for Lattice Boltzmann schemes

Uphoff, Sonja

Computational fluid mechanics has become a standard approach in many branches of engineering. Simulation of flow on the building- and infrastructure scale, however, remains very challenging and is mostly restricted to basic research at the present stage. In particular, accurate, three-dimensional, time-resolved simulation such as Large Eddy Simulation is still rarely used despite its potential. On the other hand, it is reasonable to expect a growing influence of these methods as computers become more powerful and numerical methods evolve. In the present work the Lattice Boltzmann method is chosen as a starting point to analyze simulations of flow around buildings. This approach appears to be particularly apt for such applications due to its very good scalability with respect to parallel computing. Different variants of the Lattice Boltzmann method, namely the Lattice Bhatnagar-Gross-Krook method, the Multiple Relaxation Time method, and variants of the Cascaded Lattice Boltzmann (CLB) method have been implemented and compared on the basis of standard benchmarks. Several turbulence models, such as the Smagorinsky model, the wall adapting local eddy-viscosity model, and Vreman’s model have been investigated. One focus was on the applicability of the Lattice Boltzmann method to turbulent flows, considering also the interdependence between the numerical method and the LES model. Particular attention was paid to the ability of these models to correctly reproduce turbulent shear flows. Some typical infrastructure elements have been studied and compared to wind-tunnel data. The simulations were carried out on a PC cluster and on graphics processing chips. Overall, the Lattice Boltzmann method has yielded good results for turbulent flow simulations, which is documented in several benchmarks. In particular, the results for the Factorized CLB model show for the first time for a reasonably complex benchmark, that the model performs well for turbulent flows, for which an explanation is attempted.