In order to visualise the fluid flow I cobbled together a very basic isometric renderer that layers each slice of the volume from the bottom up, alpha blending each one according to the density so as to create a passable 3D volume. The results were unexpectedly good, especially given that there are no lighting computations whatsoever.

Simulation times for a 100 x 100 x 100 volume dataset ran to about 2 seconds per frame on an AMD Athlon 3000+ processor, with rendering running at about 0.5Hz. Although not even passably real-time, processing speeds were sufficient to make tweaks and adjustments without having to wait too long to see the results.

• Turbulent flow with vorticity confinement - 567Kb (Click thumbnail to download movie. DivX 5.2.1 required)

Until I implement a geometry importer to create boundaries from arbitrary polygon meshes I'm using a simple test scene which uses a basic sphere around which the fluid can flow. While unsophisticated, I think it's quite aesthetically pleasing! I experimented with different methods of visualising the flow including pressure, divergence, curl and velocity. In the end however, the velocity field was the only one which looked any good. The visualisation of the velocity field uses colour coding to represent rates of flow (blue and green being slow, yellow, orange and red being fast). Unfortunately vorticity confinement doesn't appear to like slow-moving, completely invicid flow and introduces "creep" whereby the fluid moves by itself around the edges of the bounding box. I'm guessing this is due to round-off errors from the vorticity calculations, however the effect can be damped out by making the flow slightly viscous.

Turbulent flow without vorticity confinement - 448Kb •

The two comparisons above (one with vorticity confinement and one without) show just how much the discretisation process damps the flow. I'd like to experiment with vortex particles at a later date to see if I can render explosions. Kevin Beason has already produced some marvelous examples of this technique using his solver which are really worth checking out.

Next up will be exporting the datasets to Igneus and rendering them using the new volume renderer extension. It should be interesting to see how much more realistic the smoke appears when rendered correctly using multiple-scattered global illumination.

• Colour-coded visualisation of flow velocity - 980Kb

2nd June 2006
Adapted the interface of the 2D solver to work on a tablet PC (i.e. with a stylus and no proper keyboard input). To accomplish this, I designed a new XML-based user interface which is fully animated, skinnable and completely configurable. All the UI components (buttons, etc) can be tweaked, modified and even removed completely via a configuration file.

Download the latest version and check it out.

•   Stable solution even in high-velocity/high-viscosity flows using Semi-Lagrangian methods
•   2D and 3D simulation (2D runs at interactive framerates)
•   Vorticity confinement
•   Boundary conditions
•   Buoyancy
•   Configurable viscosity, diffusion, dissipation, time-step, vorticity and friction parameters
•   Clean XML and keyboard interface
•   Force advection, solute injection and boundary setting via mouse input
•   Imprintable bitmaps for custom boundary and solute fields
•   Pretty colours and window translucency :-)