A GPU-Accelerated TLSPH Algorithm for 3D Geometrical Nonlinear Structural Analysis

作者：He, Jiandong^*; Lei, Juanmian

来源：International Journal of Computational Methods, 2019, 16(7): 1850114.

DOI：10.1142/S0219876218501141

• TLSPH
• particle method
• geometrical nonlinear analysis
• CUDA C/C plus
• GPU accelerating

In this paper, we developed a GPU parallelized Total Lagrangian Formation of Smoothed Particle Hydrodynamics (TLSPH) algorithm for 3D geometrical nonlinear structure analysis. The code was developed using NVDIA CUDA C++. Both the TLSPH and GPU parallelization algorithms are described in detail. Compared to the traditional FEM method for structure analysis, TLSPH method is much easier to be implemented and parallelized. In addition, as a meshless based method, there is no need to mesh the domain for TLSPH method. Also, the computational cost of TLSPH is much lower than the Weakly Compressible Smoothed Particle (WCSPH) method. By introducing GPU acceleration, we have significantly improved the code performance. Two benchmark test cases for 3D geometrical nonlinear structure analysis are carried out. The simulation results are compared with analysis results and the data obtained by Abaqus, which is a popularly-used software for structure analysis based on FEM method. In order to show the efficiency of GPU parallelization, a serial code based on the same TLSPH method is also developed as a reference. Results show GPU parallelization accelerates the code obviously. In summary, the GPU parallelized TLSPH method shows the potential to become an alternative way to deal with 3D geometrical nonlinear structure analysis.

• 出版日期2019-11
• 单位北京理工大学