TheMonkey:
Searched several topics, but in particular have not found how I'm thinking about this.Intuition suggests:
- Each P, I, D should be minimized to tighten the control.
No, you tune the values for good performance, ie maximizing without bounds set by unstability and noise
- P (proportional gain - potential height of oscillations) should require a minimum multiple of the precision of measurement (if input measurement has 1 decimal of precision eg- 35.3°c, then P could be +/- 2.0°c or something, but if precision is no decimal eg- 35°c, then 2° P would not be reasonable and should be bigger. Of course P would also need a minimum based on the potential rate of change and ability to control the set point.
Have you played with PID control and seen how each parameter works in reality? I'm struggling to understand
this posting.
Typical approach is start P only, increase till marginally stable, backoff. Add D until noise becomes an issue,
backoff. Repeat till P+D are maximized without those constraints. Add I to improve accuracy and settling
(but this affects stability so P and D need re-tweaking to control oscillation).
With a non-linear system like heater control, everything is a big compromise because cooling response is not
the same as heating response. More advanced modelling of state can improve things, but your basic way to get things to work is to turn down the overall gain to regain stability at the expense of settling time - so long as
you have some I-term the system's steady state will converge.
For simple linear control like with a servomotor, tuning is pretty intuitive once you understand what each term
does.
You soon discover using PIDs that you have to handle integral-windup intelligently if there are step
changes of setpoint - the I term is only useful in settling performance and works against you when
out of equilibrium. Either have special handling to cancel integral windup, or prevent step changes
in setpoint (ie turn them into slew-rate limited ramps that don't generate unlimited wind-up).