Fuzzy logic isn't inherently risky, per se, but how it's utilized is a
different matter. Novices often get enamored with gain computation
methodologies, thinking that they've got the grail for adaptive and
self-tuning control. But all the optimal, adaptive, learning, self-tuning,
yadda yadda yadda, stuff has the same difficulties. It's really a matter of
properly constraining the algorithms so that they can only produce
valid/safe solutions. All too often, people try to use the fuzzy stuff as a
generic cure-all that gets them around the difficulty of understanding the
physics of the control problem, which could back you into an unsafe
solution. The thing is, to appropriately constrain the tuning algorithm,
you've already done a ton of analysis and already have to come up with
appropriate handling qualities criteria, performance and stability
boundaries, taking your system performance into account, yadda yadda. Given
that, you've already done the design that you're trying to design an
algorithm to achieve. So, what's the point? At the end of the day, a gain is
a gain. Your success in coming up with a viable design will depend very
little on the methodology you use to compute the gain.


"Predictor" wrote in message
oups.com...
Why is fuzzy logic "risky"?