At any rate, if the fluctuation is real and can't be avoided, then you can still improve the stability of your loop control by implementing either digital filtering at the module level of your analog input, or by creating a running average routine in your PLC program to damp the input signal.
I disagree, "
if the fluctuation is real and can't be avoided", then it is a fluctuation that must be catered for in the correct setting of the P, I, and D terms of the PID controller. The PID control algorithm has been around long enough for it to seen all sorts of process fluctuations, yet we still only have the 3 terms, applied successfully in millions of applications worldwide. I doubt there is nothing "new" that can't be accommodated by a correctly tuned PID loop.
....it is entirely possible that this level of fluctuation is due to an inherent instability in the process. If you can't achieve your desired error tolerance manually, then you'll not achieve it with a PID controller either.....
If you can achieve desired error tolerance manually, then there's no need for a PID control on it.
Let us not forget what the 3 terms of a PID controller actually do.... The I term compensates for the "under-achievement" of the P term when it is set for a stable output, and the D term compensates for the "rate of change" of either the process variable, or the error PV to SP.
The 3 terms, correctly configured, can give stable control with wildly fluctuating, undamped, feedback signals.
I'm no great expert in PID tuning, but have configured and tuned hundreds of loops, many with a "noisy" PV. I have not yet had to resort to analog input filtering or averaging.
The PID calculation has to see real PV signals, so that integral and derivative calculations are working with real process fluctuations. Hide them with filtering/damping, and you can often do away with the PID control completely.
I have also witnessed an "expert" tuning a particularly difficult cascade loop simply by watching the process for a day, making copious notes without altering anything. He returned to site on day 2 and entered new terms in the Master and Slave loops. I was in awe when the loop ran perfectly without any further adjustments. I can only assume he did lots of maths in his hotel room that night. Not once did he ever suggest the feedback needed filtering/damping/averaging etc.,