Summary: | 碩士 === 國立臺灣大學 === 化學工程學系 === 85 === This article aims to present a systematic methodology for on-line tuning
of a fuzzy logic controller.
First, we introduce the basic structure and inference
mechanisms of the crisp-type fuzzy logic controller in use.
The initial values of the fuzzy logic controller parameters are given
by choosing equally-spaced triangles as the input membership
functions, and simple control rule mapping as the control rules.
Then a method of evaluating process response performance is proposed according
to three characteristics --- percentage overshoot, rising time and
oscillation amplitude.
The performance evaluation will decide whether the tuning procedure is
necessary or not.
This evaluation will also be used as a stopping criterion
in each stage of the subsequent tuning work.
Parameters being tuned can be divided into two parts:scaling factors
and control rules.
Linquistic tuning guides for scaling factors are inferred from the analogy
of the fuzzy logic controller and the conventional PI controller.
The control rules being fired will be further adjusted according to the
discrepancy between the true response and the target response, while the later
is determined by the above-mentioned response characteristics.
Some numerical examples are supplied to demonstrate the
adequacy and efficiency
of the proposed self-tuning fuzzy logic controller.
|