An immersed boundary lattice Boltzmann method on block-structured adaptive grids for the simulation of particle-laden flows on CPUs/GPUs

We developed a highly efficient CPUs/GPUs solver for the fully resolved simulation of particle-laden flows by combining the lattice Boltzmann method (LBM) for fluid and the immersed boundary method (IBM) for fluid-structure interaction into the framework of adaptive mesh refinement (AMR). The cell-centered finite volume LBM method is adopted to guarantee the mass conservation. The boundary thickening direct force-immersed boundary method is used to capture the surface of solids to satisfy the no-slip and no-permeability boundary conditions while retaining computational simplicity. The AMR algorithm is implemented on the open-source framework AMReX, where the solids are placed only on the finest level, and different levels advance the solutions with varying steps of time, greatly reducing the computational cost and improving stability. The developed solver achieves 97.95% weak scaling efficiency on up to 128 GPUs by optimization and specific selection of BoxSize. An 18.2-fold speedup is obtained in the case of 3-level AMR meshes, compared to that on a uniform mesh, while reducing memory usage by 97.4%. Several classic cases also validated the solver, including flow past a two-dimensional fixed/oscillating circle cylinder, flow past a three-dimensional fixed sphere, and particles freely settling with stable and unstable patterns. The qualitative and quantitative results show that this highly efficient solver has good accuracy, robustness, scalability and extensibility for the complex particle-laden flows conventionally with heavy computing cost.