The Innovative CFD FAQ: Specific CFD Case Questions

This part of the Innovative CFD FAQ contains responses to the more specific types of CFD questions that have been sent in. Of course, given the nature of a FAQ document, the responses are fairly generic, but I hope they will be somewhat helpful nonetheless.

Could you write a page about [a very specific flow configuration]?

This site is intended to be, among other things, a source of general CFD information, so while I'm including examples as much as possible (given time, resources, and legal constraints), I'm not currently creating pages on request, per se. That said, if someone suggests a really cool example that I can work in (given the aforementioned constraints), I'll try to work it in.

I'm working on a [specific project/dissertation/contract] using CFD, and I'm running into problems. What am I doing wrong?

I used to be able to respond to questions like this in a very timely fashion. I would love to be able to continue to do that, but these days, work is too busy, and the traffic on the Innovative CFD site has grown as well (and the number of questions along with it). As a result, though I do occasionally answer these questions as best I can, I have not been able to keep up with them all for a while now. I hope that, by creating this FAQ, I can answer a lot of the questions up front. Maybe then the contact me form traffic will drop to a more manageable level (and the questions will be more focused too).

In grid refinement, how should I refine the grid? Is there any specific rule for this?

This is not actually a “frequently” asked question, but I thought it was a good one, so I'm posting my answer:

Ideally, when you perform a grid refinement study, you add points between the existing points in all three directions. Essentially you double the points in each direction. That way, all the points which were there before are still there, so you can directly track how the solution changes at specific points in the flow as you add resolution.

Of course, for a 3-D case, this means an eightfold increase in the number of points, and that is not always possible. This is especially true when one considers that, in general, you want to compare at least three levels of resolution. In cases where you can't afford to double resolution in each direction, it is sometimes possible to refine only in one or two directions (it becomes very tricky to analyze the results, however, so be careful).

Or, sometimes one uses methods such as replacing every two points with three. When you do that, however, you have to be careful that your solution-monitoring stations are at points which are present on all the meshes.

I'm running a and I get good agreement at low angles of attack, but at higher ones, my results are really off. Why is this, and what can I do about it?

Usually, when you see this sort of behavior, it means that the flow is separating, This is frequently a very difficult class of flow for turbulence models to get right. Even for the "standard" one- and two-equation models in production use today, it is common for either the separation point or the reattachment point (and often both) to be off.

One approach that people take to mitigate the situation is to calibrate a specific turbulence model in a specific flow solver for the specific class of flows of interest to them. This is "cheating" in a sense, but if you know that you are going to be running the same sort of flow configuration over and over again, it might make sense to do this.

If you are running a one-time simulation, then it might be worthwhile for you to look around at what others have used successfully for similar configurations. I know of one case where it was found that switching from a Menter SST model to a Chien k-epsilon model yielded greatly improved results. This is not what one would normally expect, since the SST is generally considered to be a better model, but for this specific case, it just happened that the Chien model did better. Knowing this, if I ever need to run a similar configuration, I'll probably consider tryig the Chien model, if only to see whether it makes a difference.

Other times, as has happened to me, you just have to face facts and move on. In my case, I came to realize that the 2-D airfoil simulations I was trying to run were not ever going to give good results beyond stall. I didn't have the time or the resources to run a full 3-D unsteady simulation, either, so I had to stop the work and explain everything to the customer. Not fun, but better than continuing to work on something that was never going to pay off (using the strategy we had been pursuing).



When you are finished here, you can return to the main Innovative CFD FAQ page.

Or, if you are finished looking through the Innovative CFD FAQ, return to the Innovative CFD home page and browse one of the other topics.

New! Comments

Have your say about what you just read! Leave a comment in the box below.