Hi,

I am simulating a squirrel-cage fan on STAR-CD and I am observing the influences of the CFL-rate. My conditions are: implicit calcul, k-eps model, pressure for inlet and outlet. My kinetic turbulent energy at the outlet of the fan is higher with a high CFL (10 for example)than with a low CFL (2 or 3). For the velocity at inlet it's the opposite.

I have some explanations about this results, can someone tell me whether it is plausible or not at all?

With CFL=10 the software gives results with a "macro"-view, that's to say he gives results after a crossing through ten cells. With CFL=1 it deals with a "micro-view", that is to say 1 result after one cell. Thus, we can say that the higher CFL is, the more we consider cells. As a result the turbulence is higher, and in an ideal, it would be ten times higher with CFL=10, but maybe there are interactions which decrease it.

As far as the velocity is concerned: with the "macro" view (CFL=10) we have a sum of velocity wich is lower than for the #"micro"-view (because norm of sum of vectors is shorter as sum of norms of vectors). In other words, vector speaking,the more we consider local velocity with a little CFL, the more the mean velocity will be low.

Thanks for your thoughts!!!

Hamm

How do you simulate your fan - with moving mesh or with rotating reference frame? In any case the time step should be chosen basing not only on CFL but also at RPM.

I simulate with moving mesh, with a constant rotation: 1000 rpm, and my required work is analysing influences of CFL rate. I don't take into account RPM.

Thanks!

How about: large CFL number means a larger time-step and a larger error in temporal discretisation.

I understand but there is an "engineering practice guidline" concerning choice of the time step for rotational problems - during one time step your mesh should rotate less than angular distance between two blades. So may be the problem is in the loss of accuracy, not in the CFL number.

No, because I 've already thought at this problem of angular distance between two blades. My maximal time step, that is to say where the fan rotates less than an angular distance, is such as CFL=10.5. As a result I did not calculated with more than 10, and when I spoke about high CFL, I spoke about CFL=10.

Besides STAR-CD can calculate correctly up to CFL=100.

Hrvoje is right, dude. Any PDE when discretised temporally will have truncation errors just like in spatial discretisation. Crank-Nicholson is O(dt^2) but still, errors ACCUMULATE, and proportional to the size of dt.

That CFL condition is only for stability. Try checking your .info file to see if PISO cannot converge even though the max number of corrector steps has been reached. At any time step PISO 'gives up', you should treat the results with suspicion.