Solver sensitivity in wind turbine energy calculations

[khinkelman]

Describe the bug
Hi @mwetter, I hope you and everyone are well. I came across unexpected sensitivity to solver tools and integration tolerances with a simple model calculating annual energy output of Buildings.Electrical.AC.ThreePhasesBalanced.Sources.WindTurbine. Assuming this is a not a user error, these nuanced behaviors could produce unintentional results, particularly since DASSL/CVODE with a 1e-6 tolerance is the standard recommendation.

To Reproduce
Steps to reproduce the behavior:

1. Open model 'Turbine.mo' (in Turbine.zip)
2. Run script 'SimulatePlot.mos' (in Turbine.zip), which is pre-configured with Radau (tolerance 1e-8).
3. Revise the solver/tolerance manually and simulate again for the additional results.

Files: Turbine.zip

Expected behavior
I'd expect a tolerance of 1e-6 with cvode, dassl, and radau to all produce the same results.

Screenshots
Model diagram:

Simulation results with different solvers:

Version

• Modelica Buildings Library: 12.1.0
• Modelica simulation environment: Dymola 2025x
• Operating system: Windows Server (Virtual Machine)

Additional context
Thanks for your input and advice! I'm curious if you have any insights into why this is happening and if it can be fixed by model improvements. Otherwise, if there is anything I can do to help improve documentation or the models themselves if needed, please let me know.

[mwetter]

@khinkelman : The Turbine.mo file has some strange encoding, so Dymola won't open it.

Since EWin.y starts to grow at different times, I would think the solver makes too large steps.
You could also try to set the nominal attribute, or the unbounded attribute of the state (https://specification.modelica.org/master/class-predefined-types-and-declarations.html), and it may be worth checking with the Dassault support.