[FPereira]

Hi,

I'm simulating the flow past a circular cylinder with LES, and in order to check its quality I'm writing the "instantaneous" velocity field at certain nodes (for each time step). To obtain the velocity fluctuations at these nodes I have to subtract the mean-flow velocity. However, due to the vortex shedding, the mean-flow is not constant. Can anyone tell me or give me some tips/references about the way to solve this issue. Is removing the mean-flow frequency with an fft a good/accurate approach?
Maybe the answer to this question is obvious but I'm new using LES.

Thank in advance

F.Pereira

[flotus1]

However, due to the vortex shedding, the mean velocity is not constant

Think about this statement again. The mean velocity is constant by definition.

But if you really need to separate the slow fluctuations from the fast ones for whatever reason, a high pass filtering operation could be applied.
However, I dont think that this is what you really want.

[FPereira]

Yes, the statement was not correct. I was referring to the mean-flow velocity.

However, I dont think that this is what you really want.

About your last statement.
One of my goals is to obtain the energy cascade. Then, shouldn't I consider only velocity fluctuations?

Thanks

[FMDenaro]

the key is that in LES you solve for v_bar(x,t), then you have to compute:


v_mean(x) = (1/T) Int [t0, t0+T] v_bar(x,t) dt

for T sufficiently large that is statistically meaningful.

Practically, I suggest to sample the velocity field, for examples storing the data and then post-processing the average. You need to store enough sample

[DaveyBaby]

Hi!
Following on from Denaro's comment, in order to be confident that you have made T sufficiently large and also to gain useful info about the periodic phenomena you need to first know the period of the oscillations. You can make a rough estimate of this by looking at the Strouhal number you would expect for your case (find in a journal paper) and working backwards.

[lovecraft22]

You don't need to subtract the mean field, the FFT (or whatever signal analysis you'll perform) will take care of that. Just record your velocity and feed the signal processing with that.

As already mentioned, you need to make sure your signal is recorded at a frequency that complies with the Nyquist rule and it also must be long enough for the signal analysis to return the lowest frequency you are expecting. This is however out of the original question.

[FPereira]

Thank you all for your answers. lovecraft22, I understand that for the energy cascade it might not be necessary to remove the mean field but if, for instance, your are trying to compare turbulent kinetic energy profiles with experimental data you should need to remove it, right?. I tried to remove the mean field using high-pass filters or by fft (removing the mean-flow frequencies). My point is: how accurate is this? Are there better methods to do it?


Thanks