Quote:
|
Originally Posted by phrontist
How do you determine how many error samples to include in your PID model? For the derivative portion you presumably just use the last sample (or is it common to do the average difference over a few samples?). There are well defined methods for tuning the constants in a PID algorithm, but how do you choose how far back the algorithm should look?
|
If you want the result to be zero error, you don't use a sliding finite
time window for the integral feedback. You start with zero integral
when you activate the control system and let the time integral of the
error be a running sum. The question is when to start the integral
feedback. If the proportional and derivative signals are arranged
to produce a critically damped system the system will not overshoot
at all (as a PD system). Delaying the start of the integral signal will
limit any overshoot that the integral signal injects into the system.
With regard to averaging the time derivative signal, you have to be
careful about the time delay when you do this. If you use an average
from too far back in time you could produce a sizable delay.
If you delay the signal too much you will get growing oscillations
instead of the damping that you want.
Have fun,
Eugene