search

byuflowlab / GXBeam.jl

Pure Julia Implementation of Geometrically Exact Beam Theory
MIT License
85 stars 18 forks source link

pointmass is less stable than mass #76

Closed RibeiroAndre closed 1 year ago

RibeiroAndre commented 2 years ago

Hi, Taylor. Your fix for time_domain_analysis using the new pointmass (#71) seems to work at first, but I started running some cases and I noticed it is much more unstable than using mass. It will be hard for me to give you a MWE because for simple cases it works. However, when running FSI on a flexible wing with pointmass, the solver diverges (or, more accurately, fails to converge) once the deflections become large, even if I reduce the timestep a lot. When I switch to mass, the solver is very stable again. Both approaches give me quite similar results with eigenvalue_analysis, so I don't think it's a setup issue. Also the deflections look very similar with time_domain_analysis, before the solver diverges. I'd rather use the pointmass approach, as I have lumped mass info for my model and that is slightly more accurate than using mass. But if this is very hard to fix, then I'll stick to mass. Let me know if there's something I can do on my end to diagnose what is going on. I can compare the two cases at the point where one diverges and provide you with whatever info can be useful. Thanks a lot!

taylormcd commented 2 years ago

My best guess is that using zero length elements increased the index of the DAE, resulting in the introduction of instabilities. I would need to confirm this. Unfortunately, the solution for that problem would be to reduce the index of the DAE by differentiating the governing equations.

taylormcd commented 1 year ago

I think this is effectively resolved, since zero length elements are no longer necessary in order to model point masses.