Well duh (sorry :) couldn't resist :)), It definitely is a heuristic procedure, since, try as I might, I have not been able to decisively mathematically model the process it is applied to.
But then aren't all feedback control systems heuristic by definition?

Altho' the basic model can be developed using simple mechanics theory, there are a couple of variables that just aren't consistent or measurable with my limited resources.

And the process IS non-linear,so yes alfredo the control system needed to be non-linear as well.

duh, what u are suggesting for faster control is that I include a response to the RATE of change of error between samples, which would effectively add a Derivative (D) mode control to my exisiting mode. One of the important characteristics of D-control is that it is fast and is used to compensate for the lag introduced by I-control

I still think that the basic mode here is PI, since the corrective action is proportional to the error and is integrated over time. Oscillations are not a problem because the process is fairly overdamped and I am compensating for dead-time.

I think Alfredo's argument has clarified things the most regarding my original query: I HAVE used sort of a 'fuzzy' look-up table in the algorithm, but the table hasn't been GENERATED by the algorithm itself. Rather I generated the table manually using empirical data after many hours of experimentation with the process. So it was I and not my algorithm that used fuzzy logic :)

I had read a simple description of fuzzy logic once, and thought my algo fit the description, hence the query

what i don't get right now is as duh said, there should be a "randomness" to the feedback response...how does that work? If u r going to use a fuzzy rule-set to determine the reponse, shouldn't there be some mapping somewhere between the two?

And my next doubt is, if my algorithm did somehow generate its own rule-matrix , wouldnt that qualify as AI ? Coz AI is something I dont wanna implement on a 8-bit device :D !!!


thanx,
kundi