home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.fuzzy
- Path: sparky!uunet!ukma!gatech!taco!eos.ncsu.edu!dlgerber
- From: dlgerber@eos.ncsu.edu (DARRELL L GERBER)
- Subject: Re: WHEN and WHY should I use FUZZY logic?
- Message-ID: <1993Jan24.174442.13708@ncsu.edu>
- Sender: news@ncsu.edu (USENET News System)
- Organization: North Carolina State University
- References: <C15uAt.8nH@cpqhou.se.hou.compaq.com> <1993Jan22.212749.7020@ncsu.edu> <1jq5jdINNsan@shelley.u.washington.edu>
- Date: Sun, 24 Jan 1993 17:44:42 GMT
- Lines: 61
-
- In article <1jq5jdINNsan@shelley.u.washington.edu>, chuckb@stein.u.washington.edu (Charles Bass) writes:
- |> A quick remark regarding fuzzy systems.
- |>
- |> If one doesn't need a plant model to use fuzzy logic how does
- |> one develop fuzzy rule sets/ test fuzzy rule sets and use a
- |> computer aid in generating the fuzzy rule sets?
- |>
- |> It appears that this is an important point. If I'm going to
- |> have a fuzzy Missle Guidence System and I need a plant model in
- |> order to automatically (aid) in generating a controller am I
- |> better off just using that model and standard state space
- |> analysis methods? You can't really do trial and error on such
- |> systems...
- |>
- |> If it is done purely by trial and error could similar results
- |> be had by "tuning" a PID controller (or several PID
- |> controllers).
- |>
- |> chuckb
- |>
- |> disclaimer: I hope I'm not sounding like I don't beleive fuzzy
- |> logic can develop controllers that work. This point has been
- |> made. I'm just trying to wade thru the marketing so I can
- |> understand the "big picture" a bit better.
-
-
- You make a very good point here -- this is probably one of the areas
- that needs the most work in Fuzzy Logic Control. There have been
- several effective tuning methods developed for PID control but it
- does have a big head start on Fuzzy. Part of the problem is what do
- you tune? In a previous followup to this same header I mentioned
- changing the rule base as a method of tuning a fuzzy logic
- controller (in that case on-line by using an adaptation routine).
- One could also vary the discretization functions and the shape of
- the fuzzy sets themselves (triangles, trapezoids,bell-curves, and
- sine curves have all been used). A colleague of mine tried using
- phase plots as a method of determining whether controllers are
- well tuned or not. He looked at the plots for various controllers
- from well-tuned to very poorly tuned and found a generally
- desireable pattern. Unfortunately, it didn't help him determine how
- to make a poorly-tuned controller into a well-tuned controller --
- just how to tell which is which. I think the most important thing
- missing is an understanding of exactly what effect variations in
- the rule-base, discretization functions, and fuzzy set shapes have
- on the performance of the controller. Another question is can you
- even measure the performance of fuzzy-logic controllers the same
- way as other controllers or are there better performance measures to
- be found?
-
- Just a few things to be pondered. I know I didn't answer any
- questions but maybe it helped give some better understanding into
- the state of things. Aren't new technologies great? So many
- questions, so few answers.
-
- --------------------------
-
- dlgerber@eos.ncsu.edu
- Department of Mechanical and Aerospace Engineering
- North Carolina State University
-
- ---------------------------
-