noEvent in ExteriorConvection component

[kldjonge]

In IDEAS.Buildings.Components.BaseClasses.ConvectiveHeatTransfer.ExteriorConvection, the convection coefficient switches suddenly for ceilings and floors when dT passes zero. This generates an event and can even cause chattering. Can we introduce the noEvent() operator?

if isCeiling then
    if dT > 0 then
      C = C_horz_buoyant;
    else
      C = C_horz_stable;
    end if;
  elseif isFloor then
    if ndT < 0 then
      C = C_horz_buoyant;
    else
      C = C_horz_stable;
    end if;
  else
    C = C_vertical;
  end if;

Changing the code is just changing it to:

  if isCeiling then
    if noEvent(dT > 0) then
      C = C_horz_buoyant;
    else
      C = C_horz_stable;
    end if;
  elseif isFloor then
    if noEvent(dT < 0) then
      C = C_horz_buoyant;
    else
      C = C_horz_stable;
    end if;
  else
    C = C_vertical;
  end if;

But I have little idea on the impact on reference results now. Anyway, I feel like actually forcing an additional calculation at the exact time it passes zero isn't nececarry and it is worth consisdering iit n light of simulation speed.

[jelgerjansen]

@kldjonge do you have an example where this chattering occurs?

The noEvent() operator improves simulation speed in case of chattering, but can have a negative impact on the simulation accuracy (see https://mbe.modelica.university/behavior/discrete/decay/). However, I think that the convective heat transfer equation is continuous for dT, so in combination with the smooth() operator, this should be fine.

Another solution is to follow the approach of the InteriorConvection component (spliceFunction()):

equation
    if isCeiling or isFloor then
      if linearise then
        port_a.Q_flow = if ceilingSign*dT_nominal > 0 then
              dT*coeffCeiling*abs(dT_nominal)^0.133
              else
              dT*coeffFloor*abs(dT_nominal)^0.308;
      else
        port_a.Q_flow = IDEAS.Utilities.Math.Functions.spliceFunction(
              x=ceilingSign*dT,
              pos=sign(dT)*coeffCeiling*abs(dT)^1.133,
              neg=sign(dT)*coeffFloor*abs(dT)^1.308,
              deltax = 0.1);
      end if;
    else
      if linearise then
        port_a.Q_flow = dT*coeffWall*abs(dT_nominal)^0.293;
      else
        port_a.Q_flow = sign(dT)*coeffWall*abs(dT)^1.293;
      end if;
    end if;

@Mathadon, do you have better insight into the best approach to follow? (keep it as is, use noEvent() in combination with smooth() or use spliceFunction())?

[Mathadon]

Good question! I'm not sure :p spliceFunction seems like overkill. I think that adding noEvent() would make most sense :) Perhaps even adding smooth(1, ...) resolves the state event and then no noEvent() is needed. I think that would then be the preferred option.

[kldjonge]

Suggestion to try the noEvent() approach and check the reference results if it even matters on variables that matter most?

If not, not having the events simply seems better. I think the impact on indoor temperatures would be low due to the damping effect of the thermal inertia in the roof. But indeed, in reality offcourse it will never behave discontinuous so a smoother implementation between the two constants could also work.

[jelgerjansen]

@kldjonge I'll implement this next week. However, I should check whether this implementation is actually tested (there needs to be a change in sign of dT for a floor/ceiling)