Lattice–Boltzmann simulations for complex geometries on high-performance computers

Complex geometries pose multiple challenges to the field of computational fluid dynamics. Grid generation for intricate objects is often difficult and requires accurate and scalable geometrical methods to generate meshes for large-scale computations. Such simulations, furthermore, presume optimized scalability on high-performance computers to solve high-dimensional physical problems in an adequate time. Accurate boundary treatment for complex shapes is another issue and influences parallel load-balance. In addition, large serial geometries prevent efficient computations due to their increased memory footprint, which leads to reduced memory availability for computations. In this paper, a framework is presented that is able to address the aforementioned problems. Hierarchical Cartesian boundary-refined meshes for complex geometries are obtained by a massively parallel grid generator. In this process, the geometry is parallelized for efficient computation. Simulations on large-scale meshes are performed by a high-scaling lattice–Boltzmann method using the second-order accurate interpolated bounce-back boundary conditions for no-slip walls. The method employs Hilbert decompositioning for parallel distribution and is hybrid MPI/OpenMP parallelized. The parallel geometry allows to speed up the pre-processing of the solver and massively reduces the local memory footprint. The efficiency of the computational framework, the application of which to, e.g., subsonic aerodynamic problems is straightforward, is shown by simulating clearly different flow problems such as the flow in the human airways, in gas diffusion layers of fuel cells, and around an airplane landing gear configuration. © 2020, The Author(s).