Inconsistency between version 0.10.0.dev (commit 867efc5) and version 1.0.1

[OscarEspinosaBarcenas]

Hello everybody,

I hope you are doing well. I need help with an issue in my model.

I have worked on Aviary for more than a year. My model converges using version 0.10.0.dev, commit 867efc5. However, it does not converge in version 1.0.1. It exceeds the iteration limit or reaches an infeasible solution.

I kept the same problem formulation. I updated the code for version 1.0.1 by doing things such as renaming SubsystemBuilderBase to SubsystemBuilder, updating parameter names, and loading the engine deck using prob.load_external_subsystems(), etc.

My driver settings remain unchanged. Did you make any changes to the driver or Dymos that affect convergence?

I noticed that in the new version, there is a new constraint, mission:constraints:excess_fuel_mass_capacity, that I constantly violate. Before, it was not part of the optimization. Moreover, my problem enforces MTOW at the beginning of the mission and targets the empty mass at the end. The final mass target is desirable, but previous optimizations using version 0.10.0.dev showed that the takeoff mass is not equal to MTOW. It seems that version 1.0.1 introduces a new equality constraint. I did not find documentation about changes regarding this behavior. If you can guide me, I will be grateful.

I was thinking in getting back to 0.10.0.dev (commit 867efc5) because my model converges there, but I really need optimization_history.db, which is only available in version 1.0.1.

Can you help me find the source of this problem? Let me know if you need my source code or the _out/ folder.

Thank you,

Oscar

[OscarEspinosaBarcenas]

Good morning everyone.

I think you sometimes fix problems without even reading the discussions. It was just good timing. Anyway, I updated Aviary to version 1.0.2 (not sure why it says 1.0.3.dev0) and ran the problem again. It is better! There is something you corrected that was wrong in versions 1.0.0 and 1.0.1 because it works better now. It is even faster, and I got a lower gross mass than with the previous version.

This raises questions about constraints such as Aircraft.Design.GROSS_MASS = Mission.GROSS_MASS and the way my objective function, design variables, and constraints are scaled, but this is a topic for a different discussion.

Thank you to all of you!
Oscar

[jkirk5]

Hi Oscar, sorry for the late reply. I'm glad we happened to fix your issue by coincidence. The excess fuel mass constraint was a very important addition we made around the 1.0 release because we realized there was nothing making sure that the aircraft could actually hold all the fuel the mission needed. We had all the variables needed but never actually connected them up. So this caused a lot of aircraft that previously converged, but were non-physical (needing more fuel than there was room for) to now fail.

To preserve compatibility with old models, if you set Aircraft.Fuel.IGNORE_FUEL_CAPACITY_CONSTRAINT to True, the constraint will not get added to the problem, giving you the old behavior, with the risk that your resulting aircraft is not plausible.