Viscous Layer Creation using "load script" option, possible bug

I am running into a strange issue here.  I create a 3D tet mesh within Salome 9.7, using Netgen1D2D3D and also add viscous layers.  Everything looks great, so I dump the script in order to automate the process.  When I load the script into Salome, the mesh generated is different than it was the first time, and most importantly, it has much higher nonorthogonality.  Has anyone else encountered this problem?  Attached is my script file for reference.  A workaround is to not include the "Compute" command in the script, load the file, then within the GUI, edit the mesh, open the viscous layer options without changing anything, press "OK", then "Apply and Close".  Now when the mesh computes, it is identical to running the entire way through the GUI. So based on this workaround, it appears the viscous layer in GUI vs GUI/script may be the culprit.  I appreciate any comments others may have.

Hello Brandon

I guess VL can be different in the two runs because of different 2D mesh. Could you check if it is the case.

Best regards

Edward

Edward,

Thanks for the response... I have verified that I get the same results if I run the mesh the whole way through the GUI, then clear, and re-run again, we will call this result #1.  If I then do the same thing by loading the script, I get the same results on both runs, but now different than result #1, so call it result #2.  Thus, for a given approach (pure gui or load script/gui) I have repeatability for that approach, but different results between the two approaches.  The end goal is automation of many runs with slightly different geometries, I suppose if this were not the case, I would just use my workaround mentioned in the post.

Thank you,

 

Brandon

 

 

 

Previously SMESH expert wrote:

Hello Brandon

I guess VL can be different in the two runs because of different 2D mesh. Could you check if it is the case.

Best regards

Edward

 

Edward,

You are correct about the 2D mesh being different.  I should have done this test in the first place, but I went ahead and turned off the viscous layers and I saw similar differences between running purely through the GUI and loading a script in the GUI.  My primary concern is the difference in nonorthogonality in the cells(>+/- 10) and the cell count (> +/-20%).  I wonder if there are maybe some mesh settings that are not dumped to the python script file, I will investigate some more and post any updates.

Thanks again for your comments Edward,

Brandon

 

Edward,

I was able to find the single parameter that causes the resulting meshes to differ between pure GUI and loading a script in the GUI, it is "UseSurfaceCurvature" within the NETGEN 1D-2D algorithm parameters.  When UseSurfaceCurvature is "True", the mesh between pure GUI and loading a script in the GUI are different, but when UseSurfaceCurvature is set to "False", the resulting meshes are identical.  Note that chordal error is used in both of the aforementioned scenarios.  I don't know why this is happening, but using the attached file, I compared parameter settings for both scenarios (GUI vs loading a script in GUI) and the parameters available were all the same. 

Thanks,

Brandon

Hello Brandon

Thank you for your investigation. I'll check what is wrong with UseSurfaceCurvature. I almost sure that the 2D meshes are different because of different triangulation held within OCCT geometry.

Best regards

Edward