Some very general advice for CFD...

The following comes from a couple of comments I made, and I thought it might help to have these comments not buried in a post. So...

Why CFD?

CFD will very rarely match complex flow reality. It can, but the level of effort required is considerable, so it's very rarely done. That said, it is very important to keep in mind what, exactly, you are running the CFD simulation for?

Are you trying to match complex experimental results?
Are you trying to approve / discard designs?
Are you exploring a prototype in detail?
Are you trying to validate the code?

etc.

If you want to match experimental results, you are probably going to need A LOT of nodes on the cluster, and all the RAM you can beg, borrow, or steal, because your cell count is going to be in the hundreds of millions. This is called Direct Numerical Simulation (DNS), it is difficult and expensive. And be ready to restart that simulation a dozen times while you tweak the ICs, BCs, and relaxation factors. On the plus side, setting up the physics is dead simple, since you won't have any turbulence models, or any other models that attempt to handle complexities in simple ways.

Let's talk about physics models. What is a physics model? What is a turbulence model? A turbulence model is, by definition, a simplified model of the turbulence a given flow field will experience. While the models are often very good, they are still models, built with a large number of assumptions, many of which are not correct, but which are "good enough". ALL of the models in a given CFD package are, in some way or another, a handy and acceptable simplification of the physics at play. Thus, they will never get it exactly right.

Validating CFD software usually does match reality, because the cases used to validate are exceptionally well understood, and the tweaks are in the research paper everyone reads when setting up the validation case. These are the exceptions.

Design exploration is where you want that high fidelity, but you can't afford it. You don't have the time or computing resources, so you go with the best fidelity you can and you target specific things. If I'm looking at how the flaps on a wing impact lift and drag, I'm not going to try to model the flow around the empennage, or through the engine nacelle. Those are getting some pretty basic assumptions and BCs/ICs, along with a coarse mesh, and I simply expect that the fidelity in those areas will be low. If I see evidence of some interesting fluid structures that could be the result of (for example) interactions between the flaps and the engine nacelle, then I'll refine the mesh and re-evaluate the assumptions.

And keep in mind that the reality that we know from real world or WT testing is very often a limited reality. A WT model can only host so many sensors, similarly with a full scale prototype.

So you are very rarely trying to match reality.

I'd go so far as to say that outside of academia, no one is trying to match reality. The goal is to save money and improve profitability.

2 comments