search

convexengineering / SPaircraft

Models for commercial aircraft design
http://spaircraft.readthedocs.org
25 stars 17 forks source link

persistent failure on macys #137

Open bqpd opened 5 years ago

bqpd commented 5 years ago 
Geometric Program is not fully bounded:
   Mission1.Aircraft.Engine.EngineConstants.Relax1.R has no lower bound
   Mission1.Aircraft.Engine.EngineConstants.Relax1.g has no lower bound

https://acdl.mit.edu/csi/view/convexengineering/job/gplibrary_Push_Models/189/buildnode=macys,optimizer=mosek/testReport/junit/gpkit.tests.from_paths/TestFiles/test_SPaircraft_py_mosek/

This looks like a divergent approximation (perhaps of a feasibility solve) that occurs after a few GP solves. May be fixed by switching to CCP solves?

bqpd commented 5 years ago

Regardless, if this isn't fixed soon I'll have to take SPaircraft off of GPkit PR tests until it is. :-/

1ozturkbe commented 5 years ago

Oy! I will try to fix this as soon as I can... I do think it is an issue in turbofan unfortunately...

whoburg commented 5 years ago

I assume this is a side effect of one of @1ozturkbe 's recent PRs merged on turbofan? I would much rather see a hacky fix on turbofan (e.g. add an artificial bound on R and g) along with keeping this issue open until that's sorted out, over removing this repo from GPkit tests.

1ozturkbe commented 5 years ago

I will take a look today and let you know.

bqpd commented 5 years ago

ping

On Mon, Sep 2, 2019, 11:24 Berk Ozturk notifications@github.com wrote:

I will take a look today and let you know.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/convexengineering/SPaircraft/issues/137?email_source=notifications&email_token=AALKAGEXKSR3JBUCGRF62YTQHUVYZA5CNFSM4ISEQJ7KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5WDA7Q#issuecomment-527183998, or mute the thread https://github.com/notifications/unsubscribe-auth/AALKAGEMHFPJNJEB4AGEEELQHUVYZANCNFSM4ISEQJ7A .

bqpd commented 5 years ago

I can recreate this on linux, so you might be able to on your ACDL machine

On Wed, Sep 4, 2019 at 8:54 AM Edward Burnell eburn@mit.edu wrote:

ping

On Mon, Sep 2, 2019, 11:24 Berk Ozturk notifications@github.com wrote:

I will take a look today and let you know.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/convexengineering/SPaircraft/issues/137?email_source=notifications&email_token=AALKAGEXKSR3JBUCGRF62YTQHUVYZA5CNFSM4ISEQJ7KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5WDA7Q#issuecomment-527183998, or mute the thread https://github.com/notifications/unsubscribe-auth/AALKAGEMHFPJNJEB4AGEEELQHUVYZANCNFSM4ISEQJ7A .

1ozturkbe commented 5 years ago

argh, I really don't know why this is failing... I may have to hack the model for now to get it to work... I will get it done.

1ozturkbe commented 5 years ago

Argh again, the CPU usage on my ACDL machine is literally 100% on all CPUs, so this will have to wait a little.

1ozturkbe commented 4 years ago

I haven't touched this in a while... I guess time to dig it back up. The reason this is failing is because the engine is giving completely unphysical results when coupled to SPaircraft, mainly in saying fuel burn is 1/10th the original sol'n. No idea why this is happening, because the models are identical to before. Going to PR a fix, but it will definitely not be a satisfactory solution.