Twenty years, eh? If I'm correct, high-resolution Godunov-type upwind finite volume methods were fairly new back then. Our research group uses them almost exclusively - I sit next to a guy who's working on extending a 4th-order upwind method to 3D. (Of course, the airfoil guys use centered finite difference with artificial dissipation, because that's how it's always been done, and it works. They don't need the same robustness to unexpected flow situations as we do; all they need is to be able to reliably run a steady-state case thousands of times during an optimization run.) I think the levelset method was developed around that time too.gblaze42 wrote:it's been twenty years since I worked with computational fluid dynamics. I'm curious if it's changed much?
The principles of CFD are essentially the same as they were, but the increase in computer power means we can actually do a lot more - my own research is probably going to involve using the Peng-Robinson equation of state, combined with droplet modelling equations, in 3D multiphase turbulent flow using large eddy simulation (LES). Another guy in our lab is working on a flame model with soot particle formation and volumetric radiative heat transfer. A paper I read recently shows off the versatility of the latest AUSM-based flux function by solving, if I recall correctly, the interaction of a strong shock in air with a drop of water - they had some really beautiful numerical Schlieren images, which compared rather well with the real Schlieren photos of a matching experiment...
Modelling is still necessary for almost everything when turbulence is present, but there's more detail (lots of transport equations to solve!) and less fudge because of the extra power. Sometimes models are validated against DNS nowadays...
Oh yeah, and we have a guy investigating the possibility of running some of the more highly parallelizable chunks of our code on GPUs instead of CPUs. It turns out that it works pretty well, and also (I think) that the cheapest way to get the processing power is to buy a top-of-the-line graphics card or two, because the dedicated physics GPU packages are more expensive for basically the same hardware. Unfortunately it's currently single-precision only...