connector with intertial pressure

[mwetter]

This branch is to evaluate the new fluid modeling approach from
Dirk Zimmer, Daniel Bender and Alexander Pollok presented in
"Robust modeling of directed thermofluid flows in complex networks". Japanese Modelica Conference, 2018. https://elib.dlr.de/120701/1/Modelica2018_Japan_paper_8.pdf

Development branch is issue946_inertialPressure

[Mathadon]

This indeed sounds interesting, although the fast dynamics could become a problem. I scanned the paper. Some initial thoughts before having more time to look into it in more detail:

• Fast dynamics are introduced but I'm not sure how fast they are exactly. I expect that we won't be able to use explicit Euler integration anymore, so we have to resort to implicit methods that scale worse with model size. That's a showstopper for the type of simulations that we run with IDEAS.
• If we add mass flow rate as a state, we should probably remove pressure as a state in mixing volumes to avoid a mass-damper configuration where oscillations will be produced.
• We currently have mass/pressure as a state for air. This does indeed solves some problems but at the cost of more computation time. When using mass as a state, algebraic loops can still form if there is no mixing volume between components.
• The overall number of equations increases and the complexity is offloaded to the integrator. It might then fail to converge. If that is not the case and the integrator is able to solve the equations then it would make sense to better understand why. Perhaps a different algorithm can be used to solve the algebraic loops.
• The paper mentions a constraint that loops are not allowed? I'm not sure what this implies for our models right now.
• We'd require a length and area parameter for each component that has fluid?
• To be checked further: It looks like mass is conserved explicitly from the model equations. That task is not offloaded to the integrator or algebraic loop solver. That's good.

It would be interesting to figure out why 'our' algebraic loops sometimes do not converge and fix that, rather than try to work around it.

[mwetter]

Try putting the loop breaker in the base class of the resistances, and set the default value for air volumes to be steady-state mass balance (to avoid oscillations).

• Verify that series resistances do not lead to multiple states.

Separately, the Modelica spec may have to be changed to subtract the nominal value from variables when solving algebraic loops, e.g., for air pressure, one would subtract 1E-5 Pa to avoid pressure differences that are small compared to the variable value.

[Mathadon]

If we have a clear (not too many equations and too many different types of equations) example where JModelica fails to solve an algebraic loop, I may try to debug it to see what is really going on on the numerical side. Does anybody have something like that lying around?

[mwetter]

For now, we will look into improvements to the solver and/or other model formulations
that does not require adding fast dynamics.