How to factor out a thermodynamic state record outside of a function?

[henrikt-ma]

I would like to discuss a modeling challenge encountered in the AixLib library. I am not sure whether it also exists in the MSL or other libraries dealing with multi-phase thermodynamics.

Problem

The problem boils down to the desire to factor out a computation of a ThermodynamicState record, for example, a non-trivial setState_phX call, store the result in a component of a model, and then use the component in one or more places where a continuous-time property (density, pressure, etc) is computed based on the state. Unfortunately, the ThermodynamicState record tends to include an Integer phase member, so the record cannot be equated to a function call with non-discrete-time argument expressions, and of course the arguments to setState_phX are not discrete-time.

In some cases, by sacrificing code readability and paying the price of code duplication, one can eliminate the ThermodynamicState record components and thereby comply with the variability rules. In other cases, such as PartialFlowCoefficient (https://github.com/RWTH-EBC/AixLib/blob/main/AixLib/Fluid/Actuators/Valves/ExpansionValves/BaseClasses/PartialFlowCoefficient.mo) I don't know what to do. There we have:

  input Medium.ThermodynamicState staInl
    "Thermodynamic state at valve's inlet conditions";
  input Medium.ThermodynamicState staOut
    "Thermodynamic state at valve's outlet conditions";

and if I understand the design correctly, the idea is that these are set in one place of the code, and then used in other places. Eliminating the staInl and staOut would completely ruin this design.

Ultimately, it feels to me as if the root of the problem is that an Integer component cannot be continuous-time. If it could, having a continuous-time ThermodynamicState would not be a problem either.

Possible solution: Declared non-discrete-time component variability

As a starting point for the discussion, say we would explicitly declare that the components are non-discrete-time like so:

  input continuous Medium.ThermodynamicState staInl
    "Thermodynamic state at valve's inlet conditions";
  input continuous Medium.ThermodynamicState staOut
    "Thermodynamic state at valve's outlet conditions";

or for the specific member of the state record:

  continuous Integer phase;

The rules for using something declared continuous in expressions would be similar to using expressions wrapped in noEvent(…). To avoid difficult questions about how to interpret simulation results stored to file, one could simply avoid storing the non-discrete-time variables of discrete-valued type.

Of course, there would probably be more details to sort out and the design would require effort, but what are the alternatives?

Relation to new GenerateEvents semantics?

Since we haven't implemented the new #3610 semantics for GenerateEvents = true in System Modeler, I'd also appreciate if someone with an up-to-date implementation (I am thinking of you, @HansOlsson), could shed some light on to what extent the new semantics come to the rescue here. Could it be the case that all constructions of these ThermodynamicState components already fulfill the rules of being discrete-time expressions?!

Note that this GenerateEvents = true in getPhase_ph wasn't there in 4.0.0:
https://github.com/modelica/ModelicaStandardLibrary/blob/8ae3d35c24e519cb2996cab20f3b13daf2b0c50a/Modelica/Media/R134a.mo#L2233

and that AixLib is still using MSL 4.0.0:
https://github.com/RWTH-EBC/AixLib/blob/2a0329b5899274e8d470e4e3e0e7e8a428918e49/AixLib/package.mo#L8

[HansOlsson]

First part: regarding GenerateEvents the function to look at seems to be Modelica.Media.R134a.R134a_ph.setState_phX not Modelica.Media.R134a.R134a_ph.getPhase_ph. For some reason (likely to help inlining) setState_phX has been re-written to duplicate and inline the getPhase_ph call.

Second part, if we then look at AixLib.Fluid.Actuators.Valves.ExpansionValves.BaseClasses.PartialExpansionValve.FlowCoefficient we need to figure out where it is used (I'm glad we have Find Usage... in Dymola), and as far I can see the use is e.g., in AixLib.Fluid.Actuators.Valves.ExpansionValves.Examples.ExpansionValveMassFlowRate where linearValve contains

staInl = Medium.setState_phX(port_a.p,
    actualStream(port_a.h_outflow), actualStream(port_a.Xi_outflow))
    "Thermodynamic state of the fluid at inlet condtions";

By using GenerateEvents=true for that setState_phX function it should work fine, even if contains Integer/Booleans.

[henrikt-ma]

By using GenerateEvents=true for that setState_phX function it should work fine, even if contains Integer/Booleans.

So the conclusion is that AixLib needs to switch to MSL 4.1.0?

[HansOlsson]

By using GenerateEvents=true for that setState_phX function it should work fine, even if contains Integer/Booleans.

So the conclusion is that AixLib needs to switch to MSL 4.1.0?

At least if using R134A, and similar Media. I didn't look that deep.

[henrikt-ma]

Then I don't see the need to keep this issue open anymore. A new issue should be opened if one would be specifically interested in storing continuous-valued expressions in variables.