ClosedLoopUnitIdentifier produces TimeConstant estimates that are sometimes way off (rewrite)

[steinelg]

The approach to estimte Time-constants should be rewritten to use PlantSimulator.SimulateSingle to estimate time constants, as this can be done manually but is not how the CLUI works

[steinelg]

Looking at Static_SinusDistANDSetpointStep, this is a unit test where th gain is found correclty, but the time-constant is way too high.
even though the "true" system is static, the found model has quite a large time constant.

The CLUI seems to work better if there is a setpoint change or a step change in the disturbance, but a continous "noisy" estimate.

It seems to struggle with. The global search for pid gain, has three criteria, and the two secon ones that are covariance are not worth much if there is a noisy disturbance with no setpoint changes as they are set to zero.

Thus the global search is left "just" looking at the simulated u_pid (referred to as u_pid_adjusted) and finding the gain that yields the smallest variance in this value.

[steinelg]

it might be that in some cases the closedloopunitidentifier should not return value, or at least it should give a warning or it should give more realistic indication of the uncertainty in the paramters it has found.

Look more at
Static_SinusDistANDSetpointStep

and even consider that there is not setpoint step, just a sinus-like disturbance.

[steinelg]

it is acutally easier to think of disturbance d_LF and d_HF as d_y and d_u, as d_HF is really just e= ymeas-ysetp and d_u is depends on u.

[steinelg]

this appears to be improved in version 1.3.42 after a major refactoring of CLUI