Re: FCL

On Mon, 09 Jan 2006 22:14:34 +0100, Ernst Murnleitner wrote:

>> Well, but it *is* FCL syntax. (false,0)(true.1) is "membership_function" as
>> defined on the page 18. Now, is there any good reason why
>> "membership_function" should not appear in operators AND, OR, XOR, NOT, IS?
>
> It is not documented and therefore most fuzzy programs will not support
> it, or do they?

Do you mean here FCL compilers or FCL "virtual machines" running on some
target hardware. Clearly, other compilers will not be conform. For a
virtual machine it should be no matter because a static expressions will be
resolved to a constant. It is a question what target the back end aim.

In my case it is a fuzzy machine learning system, which uses rules blocks
as training sets and builds classifiers. So the rules completely vanish in
the end.

If the target need to be a fuzzy controller, then FCL in its present state
seems ill-suited for this. It completely lacks hardware description. How
sensors are read and actuators are written? There could be many approaches
to that:

1. FCL could be extended to describe the hardware environment similar to
COBOL. This is probably the worst variant. Because if you say needed to
deal with CAN-bus, or some obscure D/A card, and FCL didn't handle it, then
what would you do?

2. FCL extended to become a universal-purpose language. This seems to be
completely unrealistic.

3. FCL embeddable in a universal-purpose language, like embedded SQL is.
This is technically possible, but quite awful.

4. FCL supplied as a library for a universal-purpose language. This is the
least effort approach, but it is becomes very difficult for small embedded
or real-time targets. (That was the reason why I used Ada, which is
portable across different hardware and OSes.) And it still does not answer
the question what to do with the rules. Similar approach use other
half-baked languages, like MATLAB/Simulink. But Simulink generates C or Ada
code, which is then compiled for the target. As for FCL it is not very
clear how rules can be compiled into code.

--
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de
.

Relevant Pages