What is the situation in a state machine where variables in the state are assigned values outside the block

[huihui11123]

model StateInOut4
  inner Real value(start=0);
  block State1
    outer output Real value;
    Real a;
  equation 
    a = value;
    //a = time;
  end State1;

  block State2
    outer output Real value;
  equation 
    value = 1;
  end State2;

  State1 state1;
  State2 state2;
equation 
  initialState(state1);
  transition(state1,state2,ticksInState()>=5,reset=true,immediate=true);
  transition(state2,state1,ticksInState()>=5,reset=true,immediate=true);

  state1.a=time;
  annotation (experiment(StopTime=7, __Dymola_Algorithm="Dassl"));
end StateInOut4;

In this model, variable state1.a is assigned a value on the outer layer.
I checked the code exported from Dymola for this model, and this equation is not within the state activation control if-statement：

when _Clocks.BaseClock_0.SubClock_1.active then
  state1.a := time;
  if (_StateMachines.state1.activeState == stateMachine_of_state1.state_state1) then 
    state1.value := state1.a;
  end if;
  if (_StateMachines.state1.activeState == stateMachine_of_state1.state_state2) then 
    state2.value := 1;
  end if;
end when;  

However, in the direct simulation results in Dymola, it has no value when the state is not activated:

If this model is exported as FMU and then imported into Dymola for simulation, variable state1.a will have a value at all times:

So what should the expected result be? I think even if assigning values on the outer layer, variables in the state should still only be calculated when the state is activated.

[huihui11123]

Dymola probably only hides the value of state1.a when state1 is inactive in the front-end display. In the background, state1.a = time is still calculated even when state1 is inactive.

I previously raised this issue only because I was uncertain whether my own tool, when implementing the state machine, needs to place the equation state1.a = time (which is assigned outside the state block) inside the activation control branch of the state.

Now it appears this issue has minimal impact, as in practical applications, people generally don't care whether variables inside an inactive state (like state1) have values or what those values might be.

[huihui11123]

Additionally, I'd like to ask: where can I find Modelica models that actually use state machines? It seems practical applications of state machines are relatively uncommon, aren't they?

[HansOlsson]

Some notes:

• The original model should use state1.a=sample(time); as you cannot use time directly in a clocked partition.
• The models are not balanced models as the equation for state1.a is outside the block state1 (so, State1 lacks an equation and StateInOut4 has one too many).
• Yes, these kind of clocked state machines in Modelica are somewhat uncommon. I believe their usability could be improved in various ways by some minor extensions, but it is unclear whether to spend resources on that.

To me an underlying assumption is that equations for the states should be inside the states, and thus the model shouldn't be legal.

However, I could not find that exact requirement. The closest is "17.3.3 Activation Handling" that has the text "When a state is suspended, its equations should not be executed, and its variables keep their values –including state-variables in discretized equations."