[magicbretzel]

Hi,

I ran an IC engine simulation and compared the wall heat transfer from Converge with the one obtained using the Woschni correlation (in GT-Power). The heat transfer predicted by Converge is 5 times smaller than the one predicted with the Woschni correlation. Is it common? I am running at fairly standard operating conditions.

Along those lines, I have a few questions regarding how to model the wall heat transfer with RANS:

1a) In "Boundary" --> "Temperature boundary condition" --> "Heat model", one of the options is to put "Global". What does this imply? Does it mean that the model specified in "Turbulence modelling" --> "Turbulence model" --> "Wall heat transfer model" is the one that is used for that particular wall?

1b) Is it the same for the option "Global" in "Near wall treatment"?

2) In Chapter 8 of the manual, the different models for wall heat transfer are described. However, it remains unclear which model should be used for regular-sized engine?

3) For a given mesh size, my yplus varies from 1 to 100. What should I do to improve the heat transfer predictions? I am using the RNG k-epsilon turbulence model, the O'kourke and Amsden wall heat transfer model, and "Standard wall function" for "Near wall treatment".

Best regards,
Guillaume

[ksrivast]

Hello Guillaume,

Are you comparing your heat transfer coefficients (HTC) results or heat flux results? HTC values are kind of arbitrary and depend on your definition. CONVERGE uses near wall temperatures to evaluate HTCs. Other codes/models can use bulk fluid temperatures. And so forth. This can cause a severe mismatch in HTC comparison. Also ensure that the same wall temperatures are being enabled for both. How do your pressure plots match with experimental data?

1. a. GLOBAL implies the model selected in turbulence.in will be used for that boundary.
1. b. Above is same of near-wall treatment.

2. Our recommendation for typical ICE cases is the O'Rourke and Amsden heat transfer model. It has worked well for us in the past. At the end of the day, these are different models, with their own assumptions, and the user is free to select between them to improve their results. As our manual notes, Angelberger has shown to consistently predict lower wall heat flux and Han and Reitz typically predicts higher heat transfer, out of our available models. So the rule of thumb is, if you feel your heat transfer is high, switch to Angelberger/GruMo. If you feel it is low, switch to Han and Reitz.

3. Ideally we would like to have y+ values in the range of 30-100 where Standard wall functions are valid and work well. For ICE cases, you'll typically see higher y+ values, esp during combustion. While the models are typically robust with slightly higher y+, you can bring down the y+ values to recommended range by employing y+ AMR or boundary embedding. For low y+ values, you can select scalable wall functions or enhanced near-wall treatment to improve results.

If there is a strong concern about wall heat transfer results and switching between our available models don't improve your results, and you feel the Woschni correlation predicts the correct/better heat flux, you can consider incorporating this correlation into a heat transfer UDF.

Hope this helps,

Sincerely,

[magicbretzel]

Hi Kislaya,

Thank you for the prompt reply. I am comparing the heat flux (in Watts) by summing up the heat transfer values from the bound*-wall.out files. I will use the enhanced wall model as well as the y+ AMR embedding, and see if that impacts the results.

[magicbretzel]

Hi Kislaya, I switched to "Enhanced wall treatment" for "Near wall treatment". However, I am now getting this warning in converge.log for a significant portion of the timesteps:

turbulence iterations= 49 error_eps= 3.9816e-03 error_tke= 5.6267e-03 omega_eps= 7.0000e-01 omega_tke= 7.0000e-01
turbulence iterations= 50 error_eps= 3.7338e-03 error_tke= 5.2653e-03 omega_eps= 7.0000e-01 omega_tke= 7.0000e-01
max iterations exceeded on turbulence

recovering .... because transport equations did not converge or energy extrapolations

I don't get these recovery messages when I run with the "Standard wall function" option. What could I try to improve the situation?

I tried increasing the maximum number of iterations for the TKE and dissipation rate transport equations from 50 to 500, but I still reach the maximum number of iterations for a significant number of timesteps, because the residuals stop going down at some point in the iterating process.

Thank you,
Guillaume

[ksrivast]

Hello Guillaume,

You could try the following :
1. Keep max_tke/eps iterations at 500 and lower the under-relaxation factors to 0.5 to see if it helps.
2. Consider lower cfl limits / time-steps. Recoveries indicate a desire by the solver for lower time-steps to be more stable and reach convergence criteria.
3. Wait for a while to see if the recoveries go away.

Let me know if you continue to face issues.

Sincerely,

[MFGT]

Originally Posted by ksrivast

2. Our recommendation for typical ICE cases is the O'Rourke and Amsden heat transfer model. It has worked well for us in the past. At the end of the day, these are different models, with their own assumptions, and the user is free to select between them to improve their results. As our manual notes, Angelberger has shown to consistently predict lower wall heat flux and Han and Reitz typically predicts higher heat transfer, out of our available models. So the rule of thumb is, if you feel your heat transfer is high, switch to Angelberger/GruMo. If you feel it is low, switch to Han and Reitz.

I made an investigation of all 4 heat transfer models recently and found GruMo-UniMORE to be very similar to O'Rourke and Amsden heat transfer model, with a reduction by ~10%.

Starting with O'Rourke and Amsden, Angelberger had ~40% less Heat Transfer, while Han & Reitz had ~50% more Heat Transfer (calculated in kW).


Investigated engine was a 1.5L DI 4 cylinder at rated Power.