PID Step Test Configuration

Question

I have a process that I would like to model in order to tune my PID controller. For this I want to perform a closed-loop step test.

What PID configuration do I need for the test? For example all the articles I can find show a PID response with no overshoot, so do I configure my PID controller for the largest gain with no overshoot? Or does the amount of gain not matter as long as there is no overshoot?

It seems like a catch-22 having to tune the PID to model the process in order to tune the PID. What am I missing here?

Note - each run of the PID controller takes me 15/20 minutes, so I would prefer to model it rather than repeatedly run tests and tweak.

Thanks!

• jonk: Usually, you want critically damped, with some overshoot, to settle in minimum time. But sometimes, you have requirements that specify a maximum overshoot. You should talk more about what your process is. There is no single right answer here.
• jonk: Also, you must have a model for your "plant" if you want to stimulate accurately. Without a good model, you cannot tune well. But to get a model takes effort. You can derive a model using impulse-response methods, if you can permit shocking your system and taking measurements. Get a book and read. There are several methods for deriving plant models (or just tuning the PID more directly) through control and response info.
• Harry Svensson: Another way of doing it would be to guesstimate the model, but that might require a high(er?) level of understanding of what you're actually doing. And it's not like your guesstimate has to be 100% correct, as long as you're in the ball park it will work.
• Chu: Model the open loop.

Answer 1

Simple heuristic methods of tuning for stable systems usually have the controller set to P only (Ziegler-Nichols)- disabling I and D, or used as an on-off controller (Relay feedback tuning methods eg. Astrom et. al). Both these methods essentially force oscillation and measure that frequency to determine part of the system characteristics. Obviously these simple robust methods are not suitable for all systems.

If extremes of the process variable are a problem you you would use a conservative setpoint rather than expecting no oscillation about the setpoint. Of course if you get too far away then the tuning may not be adequate so some judgement is still required.