GPU-Accelerated Discovery of Flow Physics in Moving-Boundary and Fluid-Structure Interaction (FSI) Problems at Massive Scales

We present a GPU-accelerated sharp-interface immersed boundary framework designed to enable direct numerical simulation of flow with moving boundaries and fluid–structure interaction (FSI) at unprecedented scales. The solver employs a sharp-interface ghost-cell method implemented entirely on GPUs using CUDA, OpenACC, and GPU-aware MPI, allowing simulations with O(10) billion grid points with highly resolved moving internal boundaries. Strong and weak scaling tests are performed on multi-GPU platforms to assess the scaling efficiency of the accelerated solver. Verification is demonstrated for flow past a range of canonical and non-canonical problems with stationary and moving boundaries, where GPU simulations achieve over 20× speedup compared to CPU implementations while preserving accuracy. High-Reynolds-number simulations reveal the formation and evolution of complex vortex structures at resolutions that were previously impractical. This capability opens new avenues for exploring unsteady flow physics in fluid–structure interaction (FSI) problems with moving or deforming boundaries, offering a powerful platform for scientific discovery across a broad spectrum of applications, including biological and bioinspired locomotion, aeroelastic flutter, vortex-induced vibrations, energy harvesting, and particulate flows.