A single‐step and simplified graphics processing unit lattice Boltzmann method for high turbulent flows. (21st March 2021)

• Record Type: Journal Article
• Authors: Delgado‐Gutiérrez, Arturo; Marzocca, Pier; Cárdenas, Diego; Probst, Oliver
• Is Part Of: International journal for numerical methods in fluids. Volume 93:Number 7(2021)
• Page Start: 2339
• Page End: 2361
• Publication Date: 2021-03-21
• Subjects: compute shaders -- fluid simulation -- GPU -- jet flow -- lattice Boltzmann Method -- LES model -- lid‐driven cavity -- OpenGL; Fluid dynamics -- Mathematics -- Periodicals; 532
• DOI: 10.1002/fld.4976
• Languages: English
• ISSNs: 0271-2091

Abstract

Abstract: In this work, a low‐computational cost graphics processing unit (GPU) lattice Boltzmann Method, coupled with the LES Vreman turbulence model is presented. The algorithm is capable of simulating low‐ and high‐turbulence flows. In contrast to the fractional‐step presented in the Simplified Lattice Boltzmann Method, the proposed work uses a single‐step approach, allowing faster computations of the macroscopic variables without losing any spatial accuracy. Inspired by a recently introduced directional interpolation method for the probability distribution functions, the macroscopic variables for different locations are computed separately, enabling an even further simplification of the steps needed to predict the following time‐step. Similar to the simplified lattice Boltzmann method, this work reduces the required memory allocation by storing only the macroscopic variables. Multiple benchmark cases are presented to compare with results reported in the literature. Excellent agreement with reports in the literature are obtained, while improving the overall computational performance of the algorithm. Abstract : A low‐computational cost graphics processing unit (GPU) lattice Boltzmann method coupled with the large eddy simulation Vreman model is presented. The algorithm is capable of simulating high‐turbulence flows. The proposed work uses a novel single‐step approach, allowing faster computations of the macroscopic variables without losing spatial accuracy. The macroscopic variables for different locations are computed separately, enabling an even further simplification of the steps needed to predict the following time‐step. This work reduces the memory allocation by storing only the macroscopic variables.