UCLA

This thesis focuses on the Lagrangian approach to fluid simulation, its parallelization, and its application in the medical imaging and simulation contexts. The fundamentals of Smoothed Particle Hydrodynamics (SPH) are analyzed, and common implementation techniques are shown. We describe our SPH implementation and show a novel approach to particle-mesh collision resolution. We also focus on the data pre-processing step, so that captured time-varying volumetric heart scans can be directly used to drive the simulation, rather than hand-crafted models. Our new mesh interpolation approach generates intermediate steps to allow stable, higher resolution simulations. Multithreading and GPU parallelism are analyzed, and a multi-CPU approach is shown, which allows the simulation to be highly scalable. We present a visualization framework, VSim, and its application to heart simulations, especially for training, education and collaboration purposes. Additionally, we show the relation between Lagrangian fluids and our previously published work on particle-based hair simulation, and we explore ultrasound volume registration methods with the purpose of enabling blood flow simulations in large volumes.