Adaptive Particle Resolution in Smoothed Particle Hydrodynamics

Abstract

In this thesis we outline the formulation, implementation and validation of Adaptive Particle Resolution (APR) in Smoothed Particle Hydrodynamics (SPH). Traditionally, SPH systems are modelled with unifrom mass distribution, through our approach we in- troduce a run time re nement and de-re nement algorithm for both 2D and 3D systems. Each particle represents a mass of uid in its local region. Particles are re ned into several particles for ner sampling in regions of complex ow. In regions of smooth ow, neighboring particles are merged. This new development o ers the capability of doing multi-resolution simulations in SPH framework. We have implemented various kinds of APR such as domain based, dynamic (based on the distance between solid and uid particles) or multiphase ows in open channels and these implementations in the SPH method provide a simple yet powerful framework for applications like o -shore structures and reactor modelling In general, the current APR implementation can be applied for simulating multiscale models or systems that develop a wide spectrum of length scales. We have clearly shown the reduction in computational cost in the APR-SPH simulations without losing the accuracy. All of our simulations were run in single machine and multiple cores (shared memory). A multi-machine approach is possible, but less e cient with the given memory archi- tecture of the simulator. For further optimizations a review of the memory allocation strategy in the simulator is required.