How do CFD'rs validate their models physically?

Logically I think the process of using CFD would be to create a model that can imitate known physical experiments and can also imitate hypothetical designs of interest.

For example I am studying impinging planar air jet heat transfer. I have a model (in Fluent) that behaves well, and several experts think it should be OK. Shouldn't I try a physical experiment with one set of dimensions and parameters, and see that the model agrees well with experiment, before I apply the model to other dimensions and parameters I want to predict without experiment?

I have not found analytic solutions for examples of my problem and don't know if there are any or where to look. I've found some published Nusselt correlations to compare my model to, but the things I've found generally don't agree with each other within a factor of two, so of course they can't all be right. Of course they all state applicability limits, and I am only speaking of applications within their limits.

I haven't seen this discussed. What's the standard practice? When do you say "finished"?

Chris,

In many industrial problems, even given 'known solutions', CFD practitioners will be cautious about expected levels of accuracy. I have noted comments from large Automotive manufacturers (no names) for instance, who take CFD results 'with a pinch of salt' ie. provides nice pictures, but accuracy +-20%, at best, a 'ballpark solution'.

In your case, as I understand your problem, you could definitely test against similar theoretical/empirical results - what springs immediately to mind is 'impingement jet driers'. A lot of work was performed in this area around 10-15 years ago (or more) & is well documented.

I hope that this helps you.

diaw...

Physical validation only makes sense in terms of coparing your solution to physical data. As you have probably observed, that's not always the easiest thing to do. It also lends itself to abuse and misuse. For example, in computing flow around an aircraft an Euler solution may give a value of lift that compares favorably to experiment. If that's all you are interested in, then great. However, the problem arises when someone takes that solution and then claims that the solver is valid for a wider range of applicability than initially shown. In that sense, every time we compute a solution for which we have no data it is probably best to consider it an extrapolation beyond the known boundaries of the solver, taken with a grain of salt until we can find enough data to push the known boundaries beyond the solution state. But that's my opinion.

Just a thought. There is quite a number of articles in the journal of food engineering, and the journal of chemical engineering studying heat transfer of impinging airflows. These papers mainly elucidate the inability of k-eps turbulence models to compute heat transfer coefficients. Hope this helps