Physically-Based Fluid Modeling using Smoothed Particle Hydrodynamics

Table of Contents


Conclusions and Future Work

Initially it was thought that the SPH computations would be the primary bottleneck in this algorithm. The intent was to implement the full set of SPH equations and then eliminate complex calculations which may not be necessary for a graphics animation. It turned out that reducing the number of particles gave sufficient speed while maintaining "fluid" motion. Unfortunately this benefit was counteracted by the lack of sufficient speed in the surface evaluation algorithm.

Overall the integration of Smoothed Particle Hydrodynamics with traditional particle systems has been shown to be a successful method for modeling fluid motion for computer graphics. It is a step beyond existing fluid modeling methods in that it can accurately model large scale movement of fluid due to the use of hydrodynamic equations of motion. It can be used as either a testing ground for CFD simulations using SPH, giving a scientist an interactive method of testing out simulation parameters before running a full blown numerical simulation. On the other hand it is also useful as a modeling tool, giving an animator a physically based method of creating animations of fluid movement.

Further research is needed into the surface evaluation problem. The use of the Cell Volume structure might be improved, as well as the load distribution between the SPH and surface calculations (since the SPH computations are running quickly, it makes sense to give some of that processing power to the surface computation). It might also be useful to take advantage of frame coherence: since the surface may not change a lot during one time step only the areas which do move need recalculation. The "continuation" method of polygonization (Bloomenthal, 1988; Wyvill, 1986b) may also be useful as a combined evaluation/generation method.

Other methods of rendering the fluid could also be researched. Volumetric rendering of non-grid based data (e.g. particles) is one possible approach. Various graphics techniques such as transparency could also prove useful.

Integrating thermal energy into this equation of state may be appropriate for modeling melting and cooling of the fluid. Thermal energy is successfully used in the molecular dynamics fluid models (Tonnesen, 1991; Terzopoulos, 1989). It would seem to be the next logical step for this algorithm. The next logical step from the multi-processing aspect of this system would be distribution of the computations. The SPH calculations and/or the surface evaluation could easily be run on an SGI Challenge (or Challenge Array), assuming a fast network and some code reorganization.

Other future work involves the definition and use of more complicated obstacles, involving rounded, or other more complex surfaces. This would open the door for creating more "real world" situations in which to model fluid movement.