An Investigation Of Parallel SPH Methods For Nonlinear Free Surface Flow Based On GPU

Investigating the hydrodynamic problems with strong nonlinear free surface phenomena is very important. Smoothed particle hydrodynamics has good adaptability to study the problems due to its meshless nature. However, the calculation efficiency is always a bottleneck for applying the weekly compressible SPH method to engineering partice. The current research work explores a method of parallel implementation of the SPH simulation by means of GPU parallel computing platform, providing a new approach to improve the calculation efficiency of SPH method.SPH method is the theoretical basis of this Dissertation and the fuction contruction method, hydrodynamic approximation method, some amendments and improved method are also explored. GPU parallel architecture and CUDA software platform are the hardware fundation of parallel architecture of this study whereas GPU execution mode, and storage mode are also discussed.Combined with the calculation characteristics of weakly compressible SPH method and the GPU execution mode, an entirely GPU based parallel computing means of SPH method is proposed. The reliability of this method is preliminary demonstrated by comparing Poisueille flow numerical simulation results.For two dimensional dam breaking problem, the particle number convergence, time step convergence, and kernel function convergence of GPU numerical simulation results are validated respectively for WCSPH method and 8-SPH method. The time consumed in the solution process is provided and the detail analysis results of GPU parallel efficiency show that both of the two methods get good acceleration effects. Moreover, the similarities and differences between the dam-breaking results of WCSPH method and 8-SPH method are also presented.For two dimensional tank sloshing problem, pure sway issues are simulated by using the two kinds of SPH methods and the calculation results of different excitation frequencies and excitation amplitudes are also given. The time consumption and acceleration performance of the two methods on GPU platform are compared. The outcomes also show that the resluts of full GPU parallel computing method are reliable and efficient.