[sidharath]

I am modelling the acoustic spectrum at near surge operation of a small high-speed centrifugal compressor numerically using LES.

I've previously modelled the same operation using DDES (SBES) and now trying to understand the impact of using explicit WALE LES model (CFX) on the acoustic spectrum and infer if there are any substantial changes in the spectra or hybrid RANS -LES models are good enough.

I've experimentally measured performance data but measured acoustic data isn't available at this point in time.
I was wondering if someone can share their experiences on quantifying the credibility of the LES results while the model is still running.

I've done timestep and grid independence study (a crude form in which 3 kinds for each spatial and temporal discretization were used) on DDES model but can't afford to do the same on LES.

The convergence is judged from the residual (rms) reaching upto to the fourth order of magnitude (10^-4) and courant number (rms) is around 10.

I was wondering if someone can introduce me to methods to check the credibility of the LES results may be like Reynolds stress (used for experimental or DNS results) or turbulent kinetic energy curve. Any ideas to check if the LES is working fine or the way it should be, would be a great help since grid and timepsteps are not studied

[FMDenaro]

I suggest to have a reading to the books of Sagaut.

Consider that in the implicit filtering formulation, there is no grid independent LES solution.
For complex geometries and flows, it is not possible to give you only one general route to assess the results. The quality of the grid is important if you want to use a wall-resolved LES. A dynamic SGS model is reccommended. If possible, energy spectra are useful to see how behaves the energy transfer.
A further check I always suggest is to perform the no-model simulation on the same grid.

[sidharath]

Thank you for the directions. I somehow missed the notification to this reply and saw it only today; my apologies.

I am following the books by Sagaut and also, I've been following the presentations made by you on LES.

I was wondering if you could please elaborate on the points of 'energy spectra' from the perspective of numerical work.

Also, I don't think I understand when you meant 'no-model' in your last line. No model as in no SGS at all? If yes, I am not using any SGS and relying on numerical dissipation.

Once again, thank you.

[FMDenaro]

Energy spectra are generally used to check LES and DNS in many test cases (but I don't know if your flow problems has some direction of homogeneity) as well as RMS, correlations, etc.
They can show if you get an artificial energy pile-up at the level of the filter cut-off caused by the numerical method that does not drain efficiently kinetic energy in the inertial range. On the other hand, they can show you also the opposite, that is an excess of dissipation introduced by the numerical method (and SGS model).
In your case you are performing an ILES and the local truncation error of your discretization should mimic the effect of an SGS model.
Have you first tested your formulation in a well known test-case?
Have you used a grid that resolve the BL or you are using wall-model bc.s?
Have you averaged the solutions over a sufficient number of samples?

However, more details about your flow problems would help to understand what variables you could you to assess the quality of the simulation.

[sidharath]

Originally Posted by FMDenaro

Energy spectra are generally used to check LES and DNS in many test cases (but I don't know if your flow problems has some direction of homogeneity) as well as RMS, correlations, etc.
They can show if you get an artificial energy pile-up at the level of the filter cut-off caused by the numerical method that does not drain efficiently kinetic energy in the inertial range. On the other hand, they can show you also the opposite, that is an excess of dissipation introduced by the numerical method (and SGS model).
In your case you are performing an ILES and the local truncation error of your discretization should mimic the effect of an SGS model.
Have you first tested your formulation in a well known test-case?
Have you used a grid that resolve the BL or you are using wall-model bc.s?
Have you averaged the solutions over a sufficient number of samples?

However, more details about your flow problems would help to understand what variables you could you to assess the quality of the simulation.

Thank you for the information and such a prompt response.
1. I'll read up more on how to compute the energy spectra from the numerical results although I vaguely remember CFX storing it for each iteration. Based on that i can see if the decay rate is 'right' and what energy spectrum is implying.

2. Yes, I am using the implicit wall modelled les, as you rightly pointed out and relying on numerical dissipation to act as sgs.

3. Unfortunately, I haven't due to the lack of time and computational resources. Although the formulation should be credible, not so sure about my grid and timestep sizes.

4. I am positive that grid captures boundary layer as I've performed a study on first cell height, number of inflation/prism as well as the rate of growth. The y+ as well as the results from Hybrid RANS-LES simulations are good. I don't think i have any reason to beleive that the bourndary layer isn't captured.

5. I think i've run the simulation long enough, around 200 revolutions of compressor wheel. Surge cycle is around 40 revolution as per the previous results and same is seen here.

I don't see a huge difference between the LES and hybrid model's results, which is causing me to suspect the integrity of les results. It could be case that hybrid model (IDDES/SBES) is doing a good job or otherwise. I should have experimental noise spectrum by next week or so and I would be in a place to better answer my question.

One other thing is for les i am using second order scheme instead of higher order which could be a source of major errors.

[FMDenaro]

Well, the problems I see:
1) if you can compute the 1D energy spectra alomng some spatial direction, I don't know exactly your details about geometry, inflow/outflow;
2) even if you compute the spectra, it is hard to compare it to typical spectra, for example we don't know if you can expect an inertial region in your problem.

I immagine you have some references to use as comparison. What variables are controlled?
What about the h+ value where the filter acts?

[sidharath]

Originally Posted by FMDenaro

Well, the problems I see:
1) if you can compute the 1D energy spectra alomng some spatial direction, I don't know exactly your details about geometry, inflow/outflow;
2) even if you compute the spectra, it is hard to compare it to typical spectra, for example we don't know if you can expect an inertial region in your problem.

I immagine you have some references to use as comparison. What variables are controlled?
What about the h+ value where the filter acts?

1. I've attached a plot showing the geometry of the compressor being modelled.

2. I do have some work published by KTH wherein they've modelled a similar compressor using LES. The energy spectrum is also published and I can see if the spectra computed in my case is any similar.

3. I am sorry, I am not sure what you mean by 'contrrled varriables'. I am using pressure inlet and mass flow outlet as boundary conditions.

4. I don't think i ever looked at roughness number values till now. Are they of importance in les or current scenario? I can find out the h+ values on monday and let you know.

[FMDenaro]

I have a doubt: you wrote are using a second order discretization, does that mean a central or upwind discretization? I wonder what is the basis of your ILES. The formulation is based on FV or FD method?