11:00 〜 13:00
[STT41-P02] Periodic boundary granular box method for initializing large scale numerical sandbox models
キーワード：numerical sandbox 、sample preparation 、periodic boundary granular box
Analog sandbox experiments have been used to model tectonic activities and study the structures created, e.g., fault formation under compression and gravitational spreading under extension. The information from such analog experiment are often limited to the deformation features in the regions near the boundaries or on the surface. To gain insights of internal stress (and strain) states, numerical simulations of sandbox experiments based on particle simulations are viable tools. With the development of high performance computing (HPC), numerical simulations of real-scale sandbox experiment using billions of particles are possible. Using real-scale numerical sandbox experiment based on discrete element method (DEM), Furuichi et. al. studied the stress state in accretionary prisms and revealed internal arching features. Owing to its controllability on particle material properties and boundary settings, real-scale numerical sandbox experiments are expected to bring more insights for the understandings of fault activities and occurrence of earthquakes.
Preparing samples for HPC-enhanced sandbox simulations is nontrivial. Free-fall is the most common technique to initialize numerical samples. For a real-scale sandbox experiment, one needs to wait billions of particles falling into the box and reaching equilibrium. In such a way, there is no explicit control of the particle distribution characteristics, such void ratio and coordination number, which are known to affect the force transmission between particles.
In this study, we present a new method for sample preparation, taking advantage of periodic boundary granular box (PG box). A PG box is a primitive assembly of particles inside a virtual box, where distribution of particles at boundaries satisfy periodic conditions. Using a simple iterative procedure, we first generate primitive PG boxes with controlled properties such as volume fraction, coordination number, and fabric isotopy/anisotropy. PG boxes are then connected and trimmed to fit arbitrary sample boundaries. In such a way, numerical samples for large scale 3D samples with complicated boundaries can be made efficiently. The convergence is checked for the failure behaviors of PG boxes with over a million DEM particles. The initiation of a real-scale sandbox sample with gentle slope is demonstrated.
Preparing samples for HPC-enhanced sandbox simulations is nontrivial. Free-fall is the most common technique to initialize numerical samples. For a real-scale sandbox experiment, one needs to wait billions of particles falling into the box and reaching equilibrium. In such a way, there is no explicit control of the particle distribution characteristics, such void ratio and coordination number, which are known to affect the force transmission between particles.
In this study, we present a new method for sample preparation, taking advantage of periodic boundary granular box (PG box). A PG box is a primitive assembly of particles inside a virtual box, where distribution of particles at boundaries satisfy periodic conditions. Using a simple iterative procedure, we first generate primitive PG boxes with controlled properties such as volume fraction, coordination number, and fabric isotopy/anisotropy. PG boxes are then connected and trimmed to fit arbitrary sample boundaries. In such a way, numerical samples for large scale 3D samples with complicated boundaries can be made efficiently. The convergence is checked for the failure behaviors of PG boxes with over a million DEM particles. The initiation of a real-scale sandbox sample with gentle slope is demonstrated.