[Gerry Kan]

Howdy Folks:


I am looking for, perhaps, a confirmation, but other comments are welcome.



At the moment I am post-processing my LES results against LDA measurements for a steady-state flow. So my first attempt is to take the mean and the arithmetic fluctuation (i.e., RMS of central difference) across all time steps. The mean velocity profiles look quite okay as expected, but my fluctuation always underpredict.


My suspicion is that the fluctuation I am calculating is only the resolved fluctuation. For me to obtain a more reasonable representation, I will also need to calculate the SGS fluctuation. As the flow is stationary, this term could be approximated by the S_ij term alone.


I am wondering if this is the correct line of thought.



Thank you very much in advance, Gerry.

[FMDenaro]

Originally Posted by Gerry Kan

Howdy Folks:


I am looking for, perhaps, a confirmation, but other comments are welcome.



At the moment I am post-processing my LES results against LDA measurements for a steady-state flow. So my first attempt is to take the mean and the arithmetic fluctuation (i.e., RMS of central difference) across all time steps. The mean velocity profiles look quite okay as expected, but my fluctuation always underpredict.


My suspicion is that the fluctuation I am calculating is only the resolved fluctuation. For me to obtain a more reasonable representation, I will also need to calculate the SGS fluctuation. As the flow is stationary, this term could be approximated by the S_ij term alone.


I am wondering if this is the correct line of thought.



Thank you very much in advance, Gerry.

Well, a congruent comparison between LES and experimental data is not simple. First of all, the experimental measurements are able to capture correctly the 3D velocity? From a general point of view, also the measure has some filtering induced by the response and resolution of the device.

Therefore it makes no sense to think that are the LES data to be improved to match the experimental measures.

[Gerry Kan]

Originally Posted by FMDenaro

Well, a congruent comparison between LES and experimental data is not simple. First of all, the experimental measurements are able to capture correctly the 3D velocity? From a general point of view, also the measure has some filtering induced by the response and resolution of the device.

Therefore it makes no sense to think that are the LES data to be improved to match the experimental measures.

Dear Professore:

I am not too worried about the uncertainty in the measuremrnts; they are point measurements and the spatial and temporal resolutions are much finer than the current grid size. The only thing, if one has to nitpik, is the fact that I only have the instantaneous velocity in two dimensions, while the third dimension is auumed to be symmetric.

I seek not to 'improve' the existing LES results, but to try to find out what could be missing, some way to compensate my simulation results without resorting to a finer grid. I thought the SGS component could fit the explaination.

You are right about the issue of comparing LES results with what is essentially 'RANS-based' measurements. My perception is that, on the turbulent spectrum, the time average is simply the zero-frequency component, and the 'fluctuation' comprises the remaining components. As such, I should be able to make some kind of comparison, when I am sure all scales, resolved and subgrid, are taken into account.

What I don't know is whether I should expect my LES results still be comparable with the measured fluctuations, or is the comparison in your opinion nonsense to begin with, since one is time-averaged, and the other 'spatially' averaged, in which case I should assert that the mean velocity is the only reliable comparison?

Gerry.

[FMDenaro]

Originally Posted by Gerry Kan

Dear Professore:

I am not too worried about the uncertainty in the measuremrnts; they are point measurements and the spatial and temporal resolutions are much finer than the current grid size. The only thing, if one has to nitpik, is the fact that I only have the instantaneous velocity in two dimensions, while the third dimension is auumed to be symmetric.

I seek not to 'improve' the existing LES results, but to try to find out what could be missing, some way to compensate my simulation results without resorting to a finer grid. I thought the SGS component could fit the explaination.

You are right about the issue of comparing LES results with what is essentially 'RANS-based' measurements. My perception is that, on the turbulent spectrum, the time average is simply the zero-frequency component, and the 'fluctuation' comprises the remaining components. As such, I should be able to make some kind of comparison, when I am sure all scales, resolved and subgrid, are taken into account.

What I don't know is whether I should expect my LES results still be comparable with the measured fluctuations, or is the comparison in your opinion nonsense to begin with, since one is time-averaged, and the other 'spatially' averaged, in which case I should assert that the mean velocity is the only reliable comparison?

Gerry.

Of course an LES solution is fully 3D, that is a quite relevant difference from your assumption in the experimental measurements.


Could you show the comparison between the RMS profiles?


However, I suggest to use the mean fields for comparisons. Are the LES sample averaged along a sufficient long period to be compared to the steady averaged experimental data?

[Gerry Kan]

Originally Posted by FMDenaro

Could you show the comparison between the RMS profiles?

Grazie Professore per il vostro aiuto, posso sendivi domani le informazioni per massaggio privato?

Cordiali saluti, Gerry.

[FMDenaro]

Si, certamente


Saluti

[Gerry Kan]

Dear Professore:

Thank you very much for your response. To summarize, you mentioned there could be inconsistencies between the LES setup and the LDA measurements, with a secondary line of investigation on using a finer grid and possibly a different SGS model (currently I am using the Deardorff one-equation model).

As a side note, I plan to extend my modelling framework to fully compressible flow, which could preclude the use of dynamic or deconvolution models, as the ones available to me are only applicable to incompressible flows.

To come up with an appropriate course of action, I need to think about it over a day or two. As the mean velocity comparison only shows minor discrepancies (U was okay, W was a little off but it can be tolerated), I am thinking if it is really the compatibility of the two data sets, in which case I should ignore the fluctuation comparison.

Once I have formulated my thoughts I will write back to you to gather your opinions. Again, thanks very much for taking the time and effort to look into the matter and I really appreciate it.

Sincerely yours, Gerry.

[FMDenaro]

Originally Posted by Gerry Kan

Dear Professore:

Thank you very much for your response. To summarize, you mentioned there could be inconsistencies between the LES setup and the LDA measurements, with a secondary line of investigation on using a finer grid and possibly a different SGS model (currently I am using the Deardorff one-equation model).

As a side note, I plan to extend my modelling framework to fully compressible flow, which could preclude the use of dynamic or deconvolution models, as the ones available to me are only applicable to incompressible flows.

To come up with an appropriate course of action, I need to think about it over a day or two. As the mean velocity comparison only shows minor discrepancies (U was okay, W was a little off but it can be tolerated), I am thinking if it is really the compatibility of the two data sets, in which case I should ignore the fluctuation comparison.

Once I have formulated my thoughts I will write back to you to gather your opinions. Again, thanks very much for taking the time and effort to look into the matter and I really appreciate it.

Sincerely yours, Gerry.

Hello,
my suggestion is to further investigate the geometry of the experimental device and check carefully if the double periodicity in you numerical model makes sense. Check, the consistence of the BCs and the scaled model (Re number). This aspect could be an issue, more than the grid refinement.
Best regards
Filippo


PS: I doubt that a full compressible model is required in your problem, are you interested in buoyancy effects?

[Gerry Kan]

Originally Posted by FMDenaro

Check, the consistence of the BCs and the scaled model (Re number). This aspect could be an issue, more than the grid refinement.

PS: I doubt that a full compressible model is required in your problem, are you interested in buoyancy effects?

Thanks I will keep that in mind.
On the PS, yes, buoyancy is of interest but not the only reason that a compressible framework is considered. I will explain this to you in private.

PS Oggi e domani devo ridipingere la casa mia. Mi scuso se non posso rispondevi subito.