graeme@cheviot.UUCP (Graeme Dixon) (04/27/85)
Since I didn't even get a single reply last time I'm repeating
this enquiry for my brother. Any replies (PLEASE) will
be forwarded to him until his group get themselves onto Usenet.
---------------------------------------------------------------
Project Development and
Formal Methodology Within AI
Recently, I have been writing systems with a particular
bent towards engineering applications (and in particular the
aerospace industry). For my sort of application, perhaps
70% of the code is true/false production rules, with little
need for fuzzy logic. In particular, I have been using two
expert system shells called "Savoir" (by ISI Ltd) and "Sage"
(by SPL Ltd) for about four months.
However, even with this relatively short experience, I
am running into management problems, and I am very
interested in developing ideas in the following areas:
1) Methods for project development, right from the initial
concept, through knowledge aquisition and design, code
and testing stages.
To a certain extent the methodology currently used for
procedural languages could be applied to these initial
stages. Several of our systems are over 500 rules in
size, and should exceed the 1000 rule barrier by this
autumn - my concern is with maintainability and docu-
mentation.
2) Formal methodology for specifying the flow of
processes, and help with project documentation to
reduce time at the prototype stage, and produce a
higher quality product.
We have made a start in this area by introducing Coding
Standards and holding regular inspection meetings.
Is anyone else thinking along the same lines?
Mark J. Dixon
British Aerospace Plc (Warton IKBS Group)
---------------------------------------------------------------
--
Graeme Dixon -
University of Newcastle upon Tyne, UK
ARPA : graeme%cheviot%newcastle.mailnet@MIT-MULTICS.ARPA
UUCP : <UK>!ukc!cheviot!graeme