[kepler123]

Hey everyone, looking for your insight on what I believe is a trivial matter, yet one that has confused me.

I'm running an LES wall resolved simulation of flow around an airfoil and have set sensors (time series output from the cell center at chosen points) along the surface to act as microphones and record the time averaged pressure at every time-step.

My question is: Is the time averaged time step what I need or do I also have to average in space along the span too? My logic tells me that there are going to be Kelvin-Helmholtz Instabilities, horse shoe vortices, etc that would make the output in the spanwise direction vary and averaging this out during the simulation would result in me killing the spanwise variations. More details on the case and what I am trying to achieve are found below.

I am studying transition and comparing it to experimental data where I have microphone data along the airfoil and the goal is to study the frequencies that are amplified and aid in this transition from laminar to turbulent flow and make a comparison with the experimental data.

For now please assume that the inflow conditions, turbulence, etc. are set up right and in accordance with the experiment, making the comparison between the two not a problem.

As asked above, I wanted to know if I need to also space average my data while I run the simulation or do I just need a time average.

The airfoil in consideration is a small span of the wing just 0.3 times the chord length with sufficient resolution of the domain in the other directions. Previous studies have shown that for my Reynolds numbers this is sufficient to resolve the 3D effects to a degree that is satisfactory considering a balance of resources used and results obtained.

P.S: Do I even have to consider the averaged pressure or the instantaneous pressure?

Any clarifications would be extremely helpful and I'm happy to answer any further questions. I looked online and could not find any help on this topic and my guide is not really helpful in this context, he brushed it off and said that time averaged is correct, did not say I do not need a space average or if I need a space average.

[FMDenaro]

First, if you have to compare to experimental measurements then you need to be homogenous with the averaging used in the experiments.


As a more general rule, the statistics extracted in LES are averaged in space along homogeneous directions (if any) and then the quantities are also averaged in time. That is done for the zero-th order statistics and also for RMS and spectra.

[kepler123]

Thanks for the quick reply Denaro.

Regarding your first statement, I totally agree that it needs to be the same as the measurements, this should have been obvious but I hadn't considered that.

The experiment just has microphones placed along a chord line (not directly behind each other for obvious reasons). So, there's no real averaging in space happening here not an average in time even though it is at a steady-state.

I will also record the data that is not averaged in time (instantaneous) even though I time average to get to steady state because the transition process with the break up of the structures is not going to be steady (some low-level fluctuations just like flapping is what I have noticed in some experiments) and this is what I think I truly need: the variations in the region once the general simulation has stabilized. Recording a time series of the averaged data at fixed points once a steady state has already been reached should ideally result in no variations and therefore a pointless time-series in my opinion.

If you or anyone else here notices any logical errors in the last paragraph, I'd appreciate a correction or some information. Thanks again for your help.

[FMDenaro]

Originally Posted by kepler123

Thanks for the quick reply Denaro.

Regarding your first statement, I totally agree that it needs to be the same as the measurements, this should have been obvious but I hadn't considered that.

The experiment just has microphones placed along a chord line (not directly behind each other for obvious reasons). So, there's no real averaging in space happening here not an average in time even though it is at a steady-state.

I will also record the data that is not averaged in time (instantaneous) even though I time average to get to steady state because the transition process with the break up of the structures is not going to be steady (some low-level fluctuations just like flapping is what I have noticed in some experiments) and this is what I think I truly need: the variations in the region once the general simulation has stabilized. Recording a time series of the averaged data at fixed points once a steady state has already been reached should ideally result in no variations and therefore a pointless time-series in my opinion.

If you or anyone else here notices any logical errors in the last paragraph, I'd appreciate a correction or some information. Thanks again for your help.

When you talk about "steady state", that is only in statistical sense, there is no steady state in turbulence.

Therefore you have to careful consider that for a large period of time you will simulate just a numerical transient. Only after some time the initial solution is statistically disregarded from the unsteady solution and you can collect the sample for performing a physically meaningful statistical analysis.
Hence, I suggest to monitor in time the global kinetic energy in your whole domain and see when the flow is really statistically steady. That appears when the kinetic energy oscillates around a mean value.
You should see a similar behaviour also for the pressure probe.

[kepler123]

Originally Posted by FMDenaro

When you talk about "steady state", that is only in statistical sense, there is no steady state in turbulence.

Yes, I agree and this is the reason I was confused about the whole averaging process because it was not making sense as it's not really a steady case and I saw it as forcing the solution to be steady.

Originally Posted by FMDenaro

Therefore you have to careful consider that you simulate for a large period of time just a numerical transient and only after the initial solution was statistically disregarded from the unsteady beahaviour you can collect the sample for performing a physically meaningful statistical analysis.

Yup, I'm currently doing this and the solution has converged satisfactorily, I ran around 4 million time steps half of them were without any averaging because I did not want to (my confusion was explained above). I have now paused and I'm contemplating what next. Ignoring the transient stages was the one thing I was certain about. How long do I wait though, now that was something I did not know.

Originally Posted by FMDenaro

Hence, I suggest to monitor in time the global kinetic energy in your whole domain and see when the flow is really statistically steady. That appears when the kinetic energy oscillates around a mean value.

This statement here clarifies all my doubts, thanks so much. I now understand that the time averaging is not really forcing it to be steady some how (which was what I disagreed with) and I now know how long to wait before I start recording data for analysis.

Originally Posted by FMDenaro

You should see a similar behaviour also for the pressure probe.

I'll look into it, thanks.

[FMDenaro]

The statistical averaging (in time and space) is performed only as a post-processing, it does not enter into the LES run that produce always a 3D unsteady solution.

[kepler123]

Originally Posted by FMDenaro

The statistical averaging (in time and space) is performed only as a post-processing, it does not enter into the LES run that produce always a 3D unsteady solution.

Oh! This makes more and more sense now. Btw, I saw your other thread where you suggested that the person read computational methods for fluid dynamics, I've got a copy from the library today. I had a CFD course, but it was not really much.

[FMDenaro]

Originally Posted by kepler123

Oh! This makes more and more sense now. Btw, I saw your other thread where you suggested that the person read computational methods for fluid dynamics, I've got a copy from the library today. I had a CFD course, but it was not really much.

Basic CFD textbooks are good but not sufficient at all to well understand and to perform a good LES.

You could find useful my notes about LES but I strongly suggest in particular the addressed references for LES

https://www.researchgate.net/publica..._Part_I_and_II

[kepler123]

Originally Posted by FMDenaro

Basic CFD textbooks are good but not sufficient at all to well understand and to perform a good LES.

You could find useful my notes about LES but I strongly suggest in particular the addressed references for LES

https://www.researchgate.net/publica..._Part_I_and_II

Thanks, I actually attended a workshop by Denaro and some others in Brussels and have some material from them, but more on examples and things to keep in mind, etc thanks for the links.