Shared memory parallelization for large scale 3D polyhedral particle simulations

Granular materials such as sands, gravels, railroad ballast, and rock are inherently highly heterogeneous and anisotropic. While they are known as one of the most widely used materials in industry, their complex behaviors remain not fully understood. Particle-based numerical methods were introduced to account for complex particle interactions yet are computationally demanding. Significant algorithmic developments have been made to enhance the computational performance, nevertheless simulations with realistic particle shape are still computationally expensive due to its complex geometry. In this study, novel parallel algorithms for polyhedral particle simulations were developed and implemented to reduce the computational cost. The parallelization study showed that the code achieved approximately 30 times speed-up with 48 cores on a LINUX machine. With this parallelized particle-based code, engineering applications were conducted: large-scale particle granular flow simulation, full-scale ballasted track simulations, and parametric study of angle of repose: The code successfully captured the runout distances of dry granular flow. This novel approach extended the capability of simulation size up to 52 million 3D polyhedral particles. In the ballast simulation, the simulations employed similar particle sizes and shapes of the ballast, as well as the full-scale geometry as the physical setup. The simulations successfully reproduced the displacement and vibration of ties in the experiment. In the angle of repose simulation, the simulations investigated the effects of input parameters on microscopic particle interactions by measuring angle of repose. The simulations demonstrated the ability to capture self-organized criticality related to natural complex system by showing the distribution of sliding mass that followed a power law relationship. The parallelized particle-based simulation extends the limits of application size by reducing computational cost. The parallelized code is successfully exploited for the study of granular material behaviors. The large-scale particle-based simulation contributes our understanding of complex behaviors of granular materials