[qls5191]

Hi, everyone. I'm running the half-channel LES code of the ABL flow for precursor generation.
Using 2400 samples, I drawed one-directional energy spectra in the streamwise direction near the top boundary to check its grid resolution.
But in only the high wave number region, the result appeared messy and noisy. (referring to the attached image) The result near the bottom didn't show this problem.
Although I tried to use the hanning window, there was no improvement.
Please let me know how to resolve the problem.

Best regards,
Hyebin Kim.

[FMDenaro]

Originally Posted by qls5191

Hi, everyone. I'm running the half-channel LES code of the ABL flow for precursor generation.
Using 2400 samples, I drawed one-directional energy spectra in the streamwise direction near the top boundary to check its grid resolution.
But in only the high wave number region, the result appeared messy and noisy. (referring to the attached image) The result near the bottom didn't show this problem.
Although I tried to use the hanning window, there was no improvement.
Please let me know how to resolve the problem.

Best regards,
Hyebin Kim.

I have some questions:


1) half-channel is for you a flat plate having periodicity in x and z directions?
2) However, you wrote "top boundary", therefore that would mean also a bottom boundary, thus I don't understand the "half-channel" meaning.
3) How can you estimate the Re_tau number?

4) From the spectra, it seems you have a fully resolved range until to the dissipative one. Is that physical or your LES solution has a lot of numerical dissipation?

5) What about your grid resolution?
6) Are you working in double precision when you computed the spectra? What if you simply compute the spectral content of a sin function, do you see a similar behavior at high wavenumbers?
7) Be aware that the spectra are averaged in time and along the spanwise direction to get a more strong statistics.

[qls5191]

Originally Posted by FMDenaro

I have some questions:


1) half-channel is for you a flat plate having periodicity in x and z directions?
2) However, you wrote "top boundary", therefore that would mean also a bottom boundary, thus I don't understand the "half-channel" meaning.
3) How can you estimate the Re_tau number?

4) From the spectra, it seems you have a fully resolved range until to the dissipative one. Is that physical or your LES solution has a lot of numerical dissipation?

5) What about your grid resolution?
6) Are you working in double precision when you computed the spectra? What if you simply compute the spectral content of a sin function, do you see a similar behavior at high wavenumbers?
7) Be aware that the spectra are averaged in time and along the spanwise direction to get a more strong statistics.

Thank you for replying.

The answers for your questions are as follows:

1-2) In the half-channel simulation, the wall(surface) was applied to the bottom boundary condition and the free-stress condition was applied to the top boundary. The boundary condition is periodic in the spanwise direction(z-dir), and the shifted periodic boundary condition was applied in the streamwise direction(x-dir).
*Munters, Wim, Charles Meneveau, and Johan Meyers. "Shifted periodic boundary conditions for simulations of wall-bounded turbulent flows." Physics of Fluids 28.2 (2016): 025112.

3) For a ABL flow, the Re was assumed as infinity becuase of very small viscous effect.

4) Actually, I don't know exactly if it's caused by physical issue or numerical dissipation.. If the problem is caused by numerical issue, I would like to know how to resolve the problem.

5) The grid resolution is delta_x=delta_z= 0.0111m, delta_y=0.00625. U_bulk is about 5m/s.

6) I'm working in double precision. And I don't understand the intend to compute energy spectra of a sin function because I want to get one-dimensional energy spectra..

7) I computed the autocorrelation averaged in time and along the spanwise direction, and the FFT of the autocorrelation was used for energy spectra.

When I drawed energy spectra in the spanwise direction, the behavior didn't occur at high wave number. (referring to attached image)

[FMDenaro]

Originally Posted by qls5191

Thank you for replying.

The answers for your questions are as follows:

1-2) In the half-channel simulation, the wall(surface) was applied to the bottom boundary condition and the free-stress condition was applied to the top boundary. The boundary condition is periodic in the spanwise direction(z-dir), and the shifted periodic boundary condition was applied in the streamwise direction(x-dir).
*Munters, Wim, Charles Meneveau, and Johan Meyers. "Shifted periodic boundary conditions for simulations of wall-bounded turbulent flows." Physics of Fluids 28.2 (2016): 025112.

3) For a ABL flow, the Re was assumed as infinity becuase of very small viscous effect.

4) Actually, I don't know exactly if it's caused by physical issue or numerical dissipation.. If the problem is caused by numerical issue, I would like to know how to resolve the problem.

5) The grid resolution is delta_x=delta_z= 0.0111m, delta_y=0.00625. U_bulk is about 5m/s.

6) I'm working in double precision. And I don't understand the intend to compute energy spectra of a sin function because I want to get one-dimensional energy spectra..

7) I computed the autocorrelation averaged in time and along the spanwise direction, and the FFT of the autocorrelation was used for energy spectra.

When I drawed energy spectra in the spanwise direction, the behavior didn't occur at high wave number. (referring to attached image)

Your flow problem is like an open-channel flow, your ABL model is nothing but a time evolving boundary layer. Therefore in my opinion is wrong to assume an infinite Re number.


You can evaluate the stress on the wall and compute the Re number. The issue is that you need a characteristic BL lenght, for example the momentum thickness. That can increase in time.



From your details I cannot say if you are resolving or not the full range of scale. I suggested to do the FFT of the sin function to check if your code solve correctly the spike in the spectral space with zero coefficienty elsewhere.


Furthermore, have you passed to the FFT exaclty the number of components define by the Nyquist frequency along x? I cannot see without the full details about the grid. And is better to work in wall-units.



If you can, at the same y compute directly the FFT along x of the u component and repeat for all the z nodes, then take the average of the coefficients. You get the same plot?


Finally, have you already tested your code in a well-know case of a planar channel flow at Re_tau=180, 390 ...??

[LuckyTran]

1) this is a developing boundary layer over flat plate, please don't ever call it a channel ever again

3) It has a Re number. The Re is not constant over the domain and differs slighly at inlet and outlet depending on the streamwise length modeled.

4) It's definitely not physical

6) that's why you should test it against a sine wave using the same sampling settings and scripts because the sine wave has known spectra spectra

[sbaffini]

It is certainly numerical, and the fact that it happens only along x is a confirmation, because that's the direction where you have a mean flow velocity. See, for example, this work.

Still, we have very little numerical or flow details here to really give any hint.