基于 GPU 并行的改进 SPH 方法对粘性流场的模拟

摘要

光滑粒子水动力学( SPH)方法对模拟破碎波问题有着良好的适应性. 基于众核架构的GPU计算平台在加速SPH方法方面有着强大的优势. 针对传统SPH方法计算效率低和计算精度差的问题,采用δ-SPH方法对腔内剪切流动、Poiseuille流动、Couette流动问题、孤立波砰击问题进行了模拟,并且提出一种基于粒子对的GPU并行计算方法. 通过比较,得到不同边界处理方法对粘性流场模拟结果的影响规律,并且研究基于粒子对和单个粒子2种不同GPU并行计算方法,对比不同计算方法的精度和CPU时间. 结果表明,采用粒子对的GPU并行方法可以使δ-SPH方法的最大加速比超过10.%The smoothed particle hydrodynamics ( SPH ) method has a good adaptability for the simulation of breaking wave problems.The GPU computing platform based on many-core architecture has a strong advantage in SPH method acceleration.In view of the low efficiency and the accuracy problem of traditional SPH method, this paper puts forward a new GPU parallel computing model based on the particle pair and improvedδ-SPH method for simulating viscosity flows such as lid-drive cavity flow, Poiseuille flow, Couette flow and solitary wave slamming. According to the comparison of different boundary handling methods, their rules on viscous flow simulation are got. Furthermore, two GPU parallel calculation methods which are respectively based on the particle pair and single par-ticle are researched, and their accuracy and CPU time are compared.The results show that the GPU parallel calcu-lation method based on particle pairs makesδ-SPH exceed 10 times of the maximum speed-up ratio.