< Back to previous page
Project
Soil Modeling, Dynamic Refinement, and Shared Memory Parallelization within SPH
The Smoothed Particle Hydrodynamics (SPH) method is an alternative to traditional mesh-based techniques such as the Finite Element Method (FEM). Due to its particle-based Lagrangian nature the method is very attractive to simulate and study complex problems involving large deformations,moving boundaries, free surfaces and multiple phases, which can be challenging for mesh-based approaches.
In this thesis we extendthe applicability of SPH in three ways: (a) the application and validation of the method for the 3D computation of cohesive soil, (b) the development of a dynamic particle refinement procedure, and (c) the parallelization of the method for shared memory computers.
The modelused for the 3D computation of soil is based on an elastic-perfectly plastic stress-strain relationship with the Drucker-Prager yield criterion. We apply the artificial stress method to avoid tensile instabilities, a numerical issue that often appears when simulating cohesive soil in SPH. The artificial stress method is extended to the 3D general stress state. The model is validated by comparing SPH results with simulations obtained with FEM.
We present a dynamic particle refinement procedure based on particle splitting. This procedure allows to dynamically increase the resolution of the discretization by replacing a refined particle with new daughter particles. The optimal separation and smoothing distance of the newly introduced particles are determined by taking into account the error introduced by the refinement in the gradient of thekernel, and by reducing possible numerical instabilities.
Finally, we develop a parallel implementation of SPH for shared memory computers. The presented approach is based on domain decomposition and space filling curves. This ensures per thread local storage of most frequently accessed data, avoids NUMA-unfriendly memory allocations, reduces data races and allows efficient calculation of symmetric inter-particle forces.</>
In this thesis we extendthe applicability of SPH in three ways: (a) the application and validation of the method for the 3D computation of cohesive soil, (b) the development of a dynamic particle refinement procedure, and (c) the parallelization of the method for shared memory computers.
The modelused for the 3D computation of soil is based on an elastic-perfectly plastic stress-strain relationship with the Drucker-Prager yield criterion. We apply the artificial stress method to avoid tensile instabilities, a numerical issue that often appears when simulating cohesive soil in SPH. The artificial stress method is extended to the 3D general stress state. The model is validated by comparing SPH results with simulations obtained with FEM.
We present a dynamic particle refinement procedure based on particle splitting. This procedure allows to dynamically increase the resolution of the discretization by replacing a refined particle with new daughter particles. The optimal separation and smoothing distance of the newly introduced particles are determined by taking into account the error introduced by the refinement in the gradient of thekernel, and by reducing possible numerical instabilities.
Finally, we develop a parallel implementation of SPH for shared memory computers. The presented approach is based on domain decomposition and space filling curves. This ensures per thread local storage of most frequently accessed data, avoids NUMA-unfriendly memory allocations, reduces data races and allows efficient calculation of symmetric inter-particle forces.</>
Date:16 Aug 2010 → 24 Nov 2014
Keywords:Smoothed Particle Hydrodynamics (SPH)
Disciplines:Applied mathematics in specific fields, Computer architecture and networks, Distributed computing, Information sciences, Information systems, Programming languages, Scientific computing, Theoretical computer science, Visual computing, Other information and computing sciences
Project type:PhD project