Keeping 'fail mesh check', how to fix?

[wgtfly]

Hi! When I am performing optimization, 'fail mesh check' will continue to appear. Only the first primal solution can be completed smoothly. After that, it will continue to evaluate obj function and 'fail mesh check'. But I don't think this is really a problem with my own grid, because when I adjust the reference point or FFD, this problem may not occur in the same case; so I want to know why the continuous 'fail mesh check' is ? How should we better solve this problem and optimize it smoothly? You can see the log for details.
logOpt.txt

[friedenhe]

I suggest you try scaler=10.0 for the shape design variables.

[wgtfly]

This will be a certain improvement. I would like to know what the principle is?

[friedenhe]

In principle, we need to scale the design variables such that their magnitude is about 1. In this case, if your shape deformation magnitude is small, like 1e-2, you need to use a scaler of 100 or 10. Otherwise, the optimizer tends to try large shape deformation magnitudes (the optimizer is not aware of any potential mesh quality issue).

[benoitpaillard]

Hi all,

From my understanding of scaler, would it make sense to implement an autoscaler that takes min and max values and scales the variable accordingly ?

[friedenhe]

Sounds like a good idea. We usually manually scale it based on the bounds, but we can definitely do it more automatically!