Slopes for the finest mesh in a 2D flat plate case about data_driven_wall_modeling HOT 3 CLOSED

Hi Jihoo,
thanks for the update. Could you show some visualizations of your results?
Best, Andre

from data_driven_wall_modeling.

Dear @AndreWeiner,

Please see the figure as follows:

The simulation was executed until 500th step for y+ = 50 and 100. As shown in the figure, the profile is already out of the range (exceeds 0.006). If the simulation kept going on, it would crash.

Therefore, I didn't try the full simulation for the 2D mapping model.

Best regards,
Jihoo Kang

from data_driven_wall_modeling.

Dear @AndreWeiner,

I realized that the 2D mapping model was wrongly set. Therefore, I changed the feature setting for wall as follows:

1. Wall slope model
   a. features : x-coordinates, y-coordinates at faces, integral average Ux, Ux at 1st face
   b. label : wall slope
2. Face slope model
   a. features : x-coordinates, y-coordinates at faces, integral average Ux
   b. label : face slope

I trained these two models with 5000 epochs, and the notebook can be found in the same location of the cloud folder.

The result Cf is given by:

The simulation crashed, and the Cf values couldn't capture the proper behavior as above. It seems to be okay in the middle of the plate, but magGradUf is too small which leads to very high nuEff at the 1st face. If we see the csv files in the simulation folder (turbulentFlatPlate_2Dmodel/yplus_100 or yplus_50), the phenomenon is basically the same with the 1D mapping model. It means that the face slopes by the ML model becomes greater than the magGradUf after a certain time step (within 7~8th step).

Hence, I concluded that the slopes at the wall and the faces from the 1D mapping model have reasonable values compared to the 2D mapping model although they are not exactly the same.

In the cloud folder, I updated what I did, and thus the folder locations are the same as before.

I hope there would be another idea to alleviate this problem like Issue #2.

Best regards,
Jihoo Kang

from data_driven_wall_modeling.