Reddit reviews Fluid Simulation for Computer Graphics
We found 5 Reddit comments about Fluid Simulation for Computer Graphics. Here are the top ones, ranked by their Reddit score.
Used Book in Good Condition
I was going to ask what range of water effects you want (waves, small ripples, splashes and droplets, topology change). Small ripples can be handled with much easier techniques than big splashes. On refreshing the thread, though, it sounds like you want behavior on the more difficult end of things. So...
Have you written a fluid simulator before: smoke, water, anything? If not, you've got your work cut out for you. Robert Bridson's Fluid Simulation for Computer Graphics is an excellent reference, but I'll give you paper links that cover most of the relevant material.
I believe LiquidSketch uses a FLIP solver. FLIP is a hybrid particle/grid method, and Animating Sand as a Fluid is the standard reference for it in graphics. SPH is of course also an option, but tends to have more stability problems, and you'll probably want a pressure step more along the lines of PCISPH or Position-Based Fluids. I'd probably recommend going with FLIP -- it's a particle-based approach, so you can capture fine features more cheaply than with a purely grid-based method, but it is much less finicky than SPH.
If you start reading the FLIP paper and feel lost, for background you might start with Real-Time Fluid Dynamics for Games, Stable Fluids, and Visual Simulation of Smoke. Practical Animation of Liquids might also be useful if you want to know about level sets and other basic issues in grid-based liquid simulation.
You could also consider using a pre-existing solver. Mantaflow is a research-oriented solver, but its code is reasonably clear and modular, especially compared to other open source fluid simulators. It's written primarily for Eulerian (grid-based) fluids, but it also has some FLIP support. You will need some fluid simulation knowledge to use it effectively. Others might know of off-the-shelf solvers that are easier to use, more flexible, or faster.
The guy that wrote this paper also wrote a book: Fluid Simulation for Computer Graphics. Which was quite usefull. I also used additional books about PDEs and linear algebra.
I didn't use a code example. I think with stuff like this the hard part is to understand the math well and then the programming is not that difficult.
I haven't looked at the paper for a long time, but I don't think it describes the projection step in detail. But it is described in the book for a similar method.
> Can you give me any more info on what types of things you simulate
There are so many different things. One example that involves physical simulation is rendering. Rendering, turning a 3d description of a scene into a 2d image, is all about simulating the pysics of light transport. Given a set of lights and surfaces you simulate how light bounces around and what a virtual observer placed somewhere in the scene would see. Another example is explosions. Cool/realistic looking explosions for movies involve simulating burning materials, fluid/gas movement, sound propagation, fracture, plastic/non-plastic deformation, the list goes on and on.
Here are some books that might get you started in the right direction
As for programming languages, you're definitely going to need to learn C/C++. Graphics applications are very resource initensive, so it's important to use a fast language. You'll probably also want to learn a couple of scripting languages like python or perl. You'll also need to learn some graphics API's like OpenGL or DirectX if you're on Windows.
I hope this helped!
90% of any sim is building the sources and frequently you have to use those sources to define the initial shape of the sim because sims are not designed to handle them - for example, all detonations are supersonic, that's what detonation means (subsonic detonations are called deflagrations). Supersonic flows are highly compressible, they result in shockwaves, which most fluid sim software does not handle AT ALL, so you're left modeling them with particle & volume sources, leaving the fluid sim to handle the subsonic aftermath.
Pyro isn't anything magic, it's really just wrappers around a standard fluid solver, so knowing how these things work internally would be enormously helpful, but that's pretty nasty from a math standpoint - Rob Bridson (one of the principle authors of Naiad/Bifrost) has a book out, but it's not for the faint of heart - https://www.amazon.com/Simulation-Computer-Graphics-Robert-Bridson/dp/1568813260 .
Fundamentally, a fluid solver is its velocity field, that's the fiddly part of it. The individual parts of a fluid solve ultimately come down to how they affect the velocity field, everything else is really just HOW they affect that field.
Disturbance is a "random" value at every voxel, turbulence is a curl noise evaluated somewhat smoothly over the fluid's domain, same for shredding, vorticity confinement and every other operation you can control (word of advice, turn off all the shape operators on the pyro solver node and add individual gas disturbance or gas turbulence nodes into the 'velocity update' input).
Pyro is fairly heavy, so is FLIP, but if you understand the underlying math, they're generally not too bad, but in my experience, most people just know which shelf buttons to push and get completely screwed when they don't work as expected.
Nice one! CFD sure can make some beautiful images/movies. How did you make this?
I got in to CFD from a computer graphics perspective originally (Robert Bridson's Fluid Simulation book) and made an interactive solver in Java which can do some fun stuff. As an example, I made this video of convection where the density is based on the color of the fluid.
And if you want to play with the solver, it's up as an applet at http://cfb.qqnoobs.com/Fluid/. Click and drag in the screen to shoot balls of fluid!