A stable and efficient semi-implicit coupling method for fluid-structure interaction problems with immersed boundaries in a hybrid CPU-GPU framework

This paper presents a stable and efficient semi-implicit coupling immersed boundary method (IBM) for simulating fluid-structure interaction problems in a hybrid CPU-GPU framework. The method enhances numerical stability by constructing and applying implicit hydrodynamic force schemes and significantly improves computational efficiency by proposing GPU-based parallel strategies. To enhance stability performance, the hydrodynamic forces obtained by the decoupled velocity correction relationships in IBM are treated implicitly as unknowns and formulated as a function of unknown structural velocities and the predicted flow field. Both the hydrodynamic forces and the equations of structural dynamics (SD) are solved simultaneously. The entire solution procedures are realized in a hybrid CPU-GPU heterogeneous parallel framework. To guarantee thread safety and minimum data transfer between CPU and GPU, the unique correspondence between computational tasks and threads is established, optimizing the overall computational performance. The accuracy, stability, and efficiency of the present method are systematically and rigorously examined by numerical simulations of a variety of rigid and deformable FSI problems in both 2D and 3D cases, providing a comprehensive understanding of its performance. It is demonstrated that the present method not only enlarges the time step by more than 50 % compared with the conventional explicit coupling method but is also suitable for FSI problems with arbitrary solid-to-fluid density ratios. Furthermore, the computational efficiency is also enhanced by 35 to 380 times. The present CPU/GPU-based semi-implicit coupling method is particularly promising in simulating both challenging rigid and deformable FSI problems, demonstrating its wide applicability.