Transported substances, such as salt, water colors and detergents, not only change fluid properties and the fluid’s natural appearance, but they also interact with surrounding materials. Naturally, physically-based simulation of transport phenomena can be an important tool for visual effects. However, even though Smoothed Particle Hydrodynamics (SPH) provide state-of-the art solutions for fluid motion in unbounded free-surface scenarios, fluid transport phenomena have seldom been addressed. Hence, the main aim of this work is to advance SPH-based fluid animation by developing efficient simulation and rendering mechanisms for fluid transport processes. Since properties of SPH are linked to field interpolations, this thesis proposes three novel sampling techniques that target an efficient reconstruction of quantity fields:
First, this work addresses issues that arise from coupling surface transport and bulk transport. Therefore, the surface is modeled by stably sampling a surface delta function with bulk particles, which results in a consistent representation of quantity fields in bulk and on surfaces. The embedded surface model fully avoids back-and-forth interpolation between bulk and surface, and, additionally, it lays the foundation for stable and symmetric boundary fluxes.
Second, the demand for high surface resolution leads to inefficiencies that are counteracted by an adaptive particle sampling within homogeneous regions in the fluid’s bulk. Time coherent sampling, which is achieved via smooth blending of particle levels, reintroduces robustness, which is especially needed for incompressible fluids.
Third, the work features a sampling mechanism for an instant high-quality ray casting of particle fields. A greedy approach reveals strict error bounds for the underlying adaptive sampling mechanism and triggers cell merging in order to ensure cache-coherent particle access.
Error predictions stabilize integration time steps during simulation and they also preserve sharp features during ray casting. In addition to the aforementioned contributions, this thesis features a fully data-parallel implementation on currently-available Graphic Processing Units, but without limiting algorithms to hardware-specific features.