A multi-resolution smoothed particle hydrodynamics with multi-GPUs acceleration for three-dimensional fluid-structure interaction problems

Abstract

This paper presents a parallel SPH of high efficiency to tackle fluid-structure interaction (FSI) problems involving large aspect ratio structures. A unified updated Lagrangian SPH algorithm is developed for modelling both the fluid and solid simultaneously, which involves employment of the Lagrangian kernel for modelling solids to avoid the tension instability problem. Considering the high computational cost of large-scale problems, we develop an adaptive multi-resolution SPH with multi-GPUs acceleration. Specifically, coupling CUDA and MPI acceleration modes are proposed, which exploit the high computational capacity of GPU devices and the cross-device parallelism properties in multi-GPUs. To enhance the occupancy of each GPU, an efficient message-passing pattern and load balancing algorithm are of interest. To address the data race in multi-resolution SPH, a combination hybrid competition mode is proposed. Through a series of optimization algorithms, the numerical efficiency of the multi-resolution SPH approach in multi-GPUs parallel computing is further promoted. After verifying the parallel efficiency of these proposed algorithms in some typical free-surface flows, the proposed model is applied to solve the FSI problem including under-water explosion problem. The results have demonstrated the effectiveness and applicability of the proposed accelerated algorithm for SPH modeling of large-scale FSI problems.