PI: Integrator implementation

You have to add anti-wind-up. If the magnitude of the error is too high, you have to stop the integral term from winding up.

Effectively this means embedding the "intState +=error" statement inside a conditional statement which tests for the absoulte magnitude of "p" and either leaves the integral term alone when "p" is too big, or resets it to the middle of its range.

-- Bill Sloman, Nijmegen

Why change it? It's doing what it's supposed to do. (I'll bet you wanted to learn something useful, didn't you?).

With a PI controller purely in the forward path, and with most plants, the control will still have some value to it (because of the integrator state) when the plant output reaches the target. Consequently, the plant will overshoot, then come back to the correct position.

There are a number of ways that you can prevent this overshoot if it is a problem.

One way is to only put the integrator in the forward path, with the proportional in the feedback path. The response will be slower, but the overshoot will be diminished or eliminated. You can do a good job of balancing the overshoot with the reduced speed by splitting up the position gain between the forward and reverse passes.

Another way is to ramp the target, rather than moving it in steps. This can significantly reduce the overshoot without slowing you down much.

Yet another way is to play clever games with feedforward, i.e. by filtering the command signal and adding the filter output to the controller's output. Feedforward won't change the stability characteristics of a linear system, but it will change (hopefully for the better) how it responds to setpoint changes.

A standard way is to stop integration when the output reaches the limitation. But still the I part leads to an overshoot. This is compensated with a fairly high P part.

Rene

You can cut the remaining wind up way down if you do this:

.. stuff .. Total = IPart + PPart

if Total > Limit then IPart = IPart - Total + Limit Total = Limit

Output Total

Decreasing what is in the integrator when you hit the limit means that the integrator has to run up once you get near the set point.

Yes, the actual code has the behavior you described. But the anti-windup doesn't reduce to zero the integral term. I reset intState when the output is near the setPoint but it's a bit trivial :)

Marco / iw2nzm

Oh, yes, I do :)

This is exactly what I tried to write in my poor English!

Ok, let's see...

mmm, interesting! I'll give it a try.

So are you saying to ramp the set point? Actually, I have a simple IIR filter on the command signal, so there aren't true steps. Maybe I can try to increase the filter action.

This is the more complex solution, so I'll try the others before.

Thank you very much, I'll write here the result of the tests! Marco / iw2nzm

This is also very interesting! Thanks Marco / iw2nzm