Two-dimensional CUDA-based regularized lattice Boltzmann method: Flow past a cylinder

Resumo

In the present work, we develop a two-dimensional, GPU-accelerated regularized lattice Boltzmann method (LBM) using CUDA to investigate the flow past a cylinder. The solver is based on the Bhatnagar-Gross-Krook (BGK) collision model with D2Q9 lattice stencils. A moments-based curved boundary condition is formulated to handle complex geometries on a uniform Cartesian grid, and the numerical stability is significantly enhanced through a regularization procedure. Lattice nodes in the vicinity of the physical curved boundary are identified as boundary nodes, where solid boundary conditions are applied. The accuracy and capability of the developed solver in handling curved geometries are verified using the benchmark problem of flow past a cylinder. The framework successfully captures key flow features such as flow separation, reattachment, wake formation, and unsteady vortex dynamics. Additionally, flow parameters including lift and drag coefficients, surface pressure distribution, and vortex shedding frequency (Strouhal number) are evaluated and compared against existing literature, demonstrating the accuracy of the method. The solver is fully implemented on the CUDA platform, leveraging the inherent locality of the LBM for high parallel efficiency. The GPU-accelerated implementation achieves significant reductions in computational time compared to its CPU-based counterpart.