higher Reynolds problem about cylinder2dflowcontroldrl HOT 1 CLOSED

Hi @ruyigolden-cudgel ,

If you change the Re through the velocity, without changing the mesh / numerical timestep, you will likely break the CFL condition and your simulation will diverge. This is likely what the NaN indicate / how they arise.

We have looked at a bit higher Res in the following study:

• https://pubs.aip.org/aip/pof/article-abstract/32/5/053605/1033993/Robust-active-flow-control-over-a-range-of?redirectedFrom=fulltext (also on ArXiv at https://arxiv.org/abs/2004.12417 )
• https://github.com/thw1021/Cylinder2DFlowControlGeneral

As you can see, the Re was changed by changing the value of nu in the code:

https://github.com/thw1021/Cylinder2DFlowControlGeneral/blob/ac5ea7fd304dedefa4489790e45c42ca9428440e/RobustDRLCylinder2DControl/Env2DCylinder.py#L157-L158

I also think we had to tune the mesh (see the different baselines / meshes for the different Re values):

https://github.com/thw1021/Cylinder2DFlowControlGeneral/tree/master/RobustDRLCylinder2DControl/simulation_base

Note that this code is not suited for going much higher in Re than what is done there. We are going quite a bit higher in Re in newer works (for example, see https://www.mdpi.com/2076-0825/11/12/359 ), but that requires using another code. Nowadays I would recommend using another system than this old code, for example see https://github.com/venturi123/DRLinFluids (described in https://pubs.aip.org/aip/pof/article-abstract/34/8/081801/2846652/DRLinFluids-An-open-source-Python-platform-of?redirectedFrom=fulltext , also available on ArXiv at https://arxiv.org/abs/2205.12699 ).

Closing the issue, please see the linked materials.

from cylinder2dflowcontroldrl.