At 11:35 05/18/99 +0200, Tjaart van der Walt wrote:
>Stefan Sczekalla-Waldschmidt wrote:
>> Tjaart van der Walt wrote:
>> > Good advice. I have noticed another thing that can make your
>> > system much more immune to the low-frequency I-oscillations :
>> > Reset the I-value to zero whenever P=0.
>>
>> looks like a good Idea - BTW did you try to optimize your PID-paramters
>> according to Ziegler-Nichols ?
><GRIN> I optimised them according to Tjaart-Experiment!
>
>The resetting of I compensates (among other things) for
>the digital nature of the calculations. Running 16 bit
>PID values in a PIC is not feasible, so I pulled a few
>tricks out of the hat. Another, is to only accumulate
>the I value every 10-50 (settable) samples. Effectively
>this divides the I term which is always much smaller
>than the P or D terms.

actually, there are two more possibilities to handle the i part, especially
for slow (eg. temperature) processes where it is for long intervals of
ramping (like heating or cooling) far away from the setpoint: you can limit
it and/or you can reset it to 0 whenever the difference is too big (ie.
you're positive that you're not in a deviation, but in a ramp).

if the i part is already all the way to the max when the system actually
starts regulating after spending a long time in a ramp approaching the
setpoint, that's an almost sure recipe for oscillations.

ge