Abstract

Abstract When a traditional Weakly-Compressible Smoothed Particle Hydrodynamics (WCSPH) model is used to simulate free surface flow with a large Reynolds number, an unstable numerical calculation due to high random pressure oscillations will result, while an accurate pressure field is of vital significance for simulating violent fluid-structure interactions. Riemann-based SPH and Delta-SPH are widely used to solve this problem. In this paper, to enhance computational efficiency, the SPH method is implemented on a General Processing Unit (GPU) platform using Compute Unified Device Architecture (CUDA). Parallelized SPH programs including the standard SPH method, Riemann-based SPH and Delta-SPH are verified by a dam break model with large Reynolds number and violent deformation of free surfaces. The results show that all SPH methods can vividly reflect the whole process of splashing, rolling and backward jet flow; both the Riemann-based SPH and the Delta-SPH methods are effective in alleviating the problem of inhomogeneous pressure distribution in the simulation process; the Riemann-based SPH method has better stability even with relatively large particle spacing, and it has higher accuracy in simulating impact pressure. When the number of particles reaches 100,000, compared with a single-thread Central Processing Unit (CPU) implementation, the speedups obtained with NVIDIA Titan V with high computing cores and Quadro K2200 with low computing cores are thousands and hundreds, respectively.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call