An efficient implementation of the graphics processing unit-accelerated single-step and simplified lattice Boltzmann method for irregular fluid domains

Abstract

In this article, an efficient implementation of the graphics processing unit (GPU)-accelerated single-step and simplified lattice Boltzmann method for curved (irregular) fluid domains (ISSLBM) is presented, allowing the algorithm to predict the macroscopic flow variables in realistic scenarios, such as the wind flow influenced by complex terrains. The fluid domain is approximated and reconstructed with two- and three-dimensional non-uniform rational B-splines functions, allowing customized refinements for desired regions. The model accuracy is investigated by conducting a two-dimensional flow around a circular profile for cases with low Reynolds numbers (Re = 20 and 40). Furthermore, the model is also used to simulate a highly turbulent wind flow (Re = 10 × 106) around the Bolund hill, located in Denmark. Numerical and experimental results reported in the literature are directly compared with the results from the ISSLBM algorithm, producing results with excellent agreement in all metrics. The computational performance is also analyzed, showing that the GPU-accelerated ISSLBM is significantly faster than other simulations reported in the literature.