Backend issues with ThermalSeparation library

[Per Östlund]

Most of the ThermalSeparation library now flattens using the new frontend (-d=newInst), but the majority of the models fail in the backend due to various issues.

Most of the failing models gives errors such as (this one is from Examples.Absorption_IdealGases):

BackendDAEOptimize.makeEquationToResidualExp failed to transform equation: column1.balanceEquations.bool_eps[1] = false to residual form!

and seem to finally fail completely with the error:

Internal error function createNonlinearResidualEquations failed

The library can be found at: ​https://github.com/thom-marx/ThermalSeparation
The current status of the library can be found at: ​https://libraries.openmodelica.org/branches/newInst/ThermalSeparation/ThermalSeparation.html

[Karim Adbdelhak]

The problem is, that there are if equations which get processed as residual equations and some equations which cant get processed as residual equations (boolean equations), in the same condition. A minimal example is the following:

model minimal_bool_if
Boolean b;
Real x;
equation
if time>0.5 then
 b = false;
 x = cos(time);
else
 b = true;
 time = x;
end if;
end minimal_bool_if;

the second equation of the else part, time = x, will be reformed to a residual equation, because x is not on the lhs. Maybe we generally want to make a better approach here? The problem here is, that now all other equations in this if equation will get reformed to residual form, which is not possible. I changed that in my local branch such that each equation pair will get processed individually, but that requires a uniform order inside the if equation to always work as intended.

For the given model i am not quite sure if this still is the problem since it is quite complex, the new error message is as follows:

Error: Internal error function createNonlinearResidualEquations failed
[/home/karim/workspace/OpenModelica/OMCompiler/Compiler/SimCode/SimCodeUtil.mo:3335:7-3335:48:writable] Error: Internal error createOdeSystem2 failed for Generic Jacobian via directional derivatives

Next i will try to generally skip the simplification and see if that works.

[Francesco Casella]

Replying to Karim.Abdelhak:

What about asking the library developers to rewrite

model minimal_bool_if
  Boolean b;
  Real x;
equation
  if time>0.5 then
    b = false;
    x = cos(time);
  else
    b = true;
    time = x;
  end if;
end minimal_bool_if;

as

model minimal_bool_if
  Boolean b;
  Real x;
equation
  b = if time > 0.5 then false else true;
  x = if time > 0.5 then time else cos(time)
end minimal_bool_if;

?

[Karim Adbdelhak]

Replying to casella:

Willi had the same idea, maybe it is a good idea to give out the general advice to avoid mixing types in if cases, or to avoid multiple equations in if cases in general.

I will try it with one small model, since this is only a sub-problem. Mainly the residuals for the jacobian of non-linear systems cannot be generated, quite possibly there is more to it.

[Francesco Casella]

I mean, if a variable-structure model has some structure, one should try to help the tool exploit it by making it explicit in the Modelica source code.

Ideally, the back-end should be able to do that automatically, but maybe that's asking too much in general.

[Francesco Casella]

Added Thomas Marx-Schubach and Andreas Heuermann in cc:

[Francesco Casella]

@thomas.marx, in your e-mail you mention that ThermalSeparation.Components.HeatExchanger.CcounterFlowHeatExchanger is a single component, so it cannot be simulated because it's singular. In this case, please remove the experiment annotation from the source code, because it doesn't make sense.

[Willi Braun]

The model ThermalSeparation.Examples.Absorption_IdealGases starts to simulate with the following script and the changes in the the ​PR2.

loadFile("../ThermalSeparation/package.mo");
getErrorString();

setCommandLineOptions("-d=newInst,-NLSanalyticJacobian");
getErrorString();

simulate(ThermalSeparation.Examples.Absorption_IdealGases);
getErrorString();

Also we have to disable the analytical jacobian generation -d=-NLSanalyticJacobian for some reason. I will create a separate ticket.