Fluent - incorrect inlet turbulence intensity (LES of a jet)
 

Hi there,


I've encountered a serious problem with Fluent during LES of the non-reacting jet.



Params:
Re=23 750
nozzle diam=4.6 mm
velocity profile defined at the inlet (Umax=100 m/s, Umin=3.5 m/s)
inlet perturbation - spectral synthesizer with Ti=0.1%
solver: press-vel coupling - SIMPLE, momentum - central differencing, time-integration - bounded second order implicit


I expect to get RMSE velocity at the inlet: 0.1% * 75 m/s (Umean) approx. 0.075 m/s however I get ~3 m/s!!!



I decided to turn off the inlet perturbation and I get the same RMSE value ~3 m/s!


It seems like Fluent calculates Ti using some fraction of inlet velocity apart from specified Ti value... This is completely unphysical and leads to false results for the whole class of flow problems!!! What the hell? Could someone explain it to me please?


Best regards,
Jakub

https://ibb.co/Mg1M2W0
https://ibb.co/G2gwP6g
https://ibb.co/gdRQjTq

Try with the vortex method and note the differences.

NOPE. After 500 time steps with vortex method, no difference, Ti~3% of the mean flow velocity at the inlet plane. This is too much.


look at my setup and results:

https://ibb.co/nsy6yW5
https://ibb.co/BrxRKZH

I can't understand that, if this is true bug then Fluent goes into serious troubles...

I have used both methods up to Fluent version 15, yet I used the full stresses at inlet.

While the spectral synthesizer had its problems, none of the methods failed in correctly specifying the inlet.

How long have you taken the average running before evaluating the result?

Specification of the Reynolds stresses may do the trick since doing so an explicit specification of Ti is omitted.



But wait a minute! OMG!!! I read manual once again and at the very bottom it says: "The turbulence intensity value specified at a velocity inlet for LES, as described in Section 12.20.4, is used to randomly perturb the instantaneous velocity field at the inlet. It does not specify a modeled turbulence quantity."


OK, it answers my question partially, so it appears that the Ti value is modelled. So, my question now is: how to specify Ti for mean flow, not for perturbation at the inlet as BC then, instead of modelling it?


BTW sbaffini thank you for your interest into my problem.

Please consider reading an updated version of the manual:

https://www.sharcnet.ca/Software/Ans...ec_les_bc.html

Also, you are misunderstanfing the meaning of that part of the manual.

Yet, this doesn't mean you aren't right (i.e., that specification method you are using might be wrong somehow), but have you reset the statistics before starting the new averaging and for how many iterations have you run the statistics before checking the contours?

"Also, you are misunderstanding the meaning of that part of the manual." - it may be, however I can see from the results with different perturbations method and no perturbed inlet, that Ti is almost the same (~3% of mean velocity) at the inlet apart from the value 0.1% specified in inlet menu which applies only to small perturbations of velocity profile using one of the available methods. What I really don't understand is that why Fluent models turbulence intensity at the inlet in LES instead of getting it explicitly like in RANS?

"but have you reset the statistics before starting the new averaging" - yes

"for how many iterations have you run the statistics before checking the contours" - (let say: L - domain length, U- jet mean velocity, t=L/U - time in which fluid element reaching outlet) so statistics were collected for 2*t, after initial 2*t when flow field was established

Now, I am looking for a method to specify Ti explicitly, similarly as the velocity profile which I defined at the inlet.

Problem solved. abenhadj from ANSYS Student Community found that unbouded Central Differencing scheme leads to oscillations at the inlet, hence increased velocity RMS.


BR,
Jakub

This doesn't really explain anything, because my experience is that using the full stresses at inlet you get the correct rms at inlet, with both spectral synthesizer and vortex method, all of this using the unbounded central scheme (however, don't use the spectral synthesizer).

Also, if you evaluate the rms at inlet you should actually get what you set as inlet, which is not affected by the computation. Moreover, your pictures didn't seem to show any appreciable fluctuation.

If using the bounded central scheme worked for you, then probably you had other issues elsewhere. This might either be your setup (CFL, grid, etc.) or the inlet method not working as expected or the way you postprocess things.

Indeed, as a side note: you are just specifying a constant % intensity level on the whole section and, somehow arbitrarily, you decided that it is the maximum RMS level that should match your specification. Why not the minimum or some weighted average? Have you tried evaluating an area/mass weighted average of the rms over the whole inlet section?

As an additional side note: I wouldn't blame, in general, Fluent if the inlet turns out to be wrong when specifying fluctuations with a constant % turbulent intensity level. The way they were developed assume at least a k and epsilon profile.

"Moreover, your pictures didn't seem to show any appreciable fluctuation." - Huh?! 30 x more is not appreciable fluctuation?

"If using the bounded central scheme worked for you, then probably you had other issues elsewhere. This might either be your setup (CFL, grid, etc.) or the inlet method not working as expected or the way you postprocess things." - at first you are saying that I wrong because my pictures shows no big fluctuation according to you, now you are saying that CFL or grid is not OK and this is reason for big fluctuation at inlet?!

"Indeed, as a side note: you are just specifying a constant % intensity level on the whole section and, somehow arbitrarily, you decided that it is the maximum RMS level that should match your specification. Why not the minimum or some weighted average?" - man, what is unclear to you, can't you see that when taking e.g. surface average of the big RED region, coloured by value 2.5m/s, I somehow will get some other quantity?!

"As an additional side note: I wouldn't blame, in general, Fluent if the inlet turns out to be wrong when specifying fluctuations with a constant % turbulent intensity level. The way they were developed assume at least a k and epsilon profile"- this clearly indicatesnumerical/discretization issues, I've tried several meshes 1, 4 and 12 millions elements...

I have feeling that you clearly can't see the problem. While I and admin from ANSYS forum were able to spot that and find the solution.

You and admin really seem LES experts, especially in Fluent... good for you then.

Cheers