Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!srcsip!falcon!funk
From: funk@falcon.SRC.Honeywell.COM (Harry Funk)
Newsgroups: comp.ai
Subject: Re: Model based diagnosis products/work
Message-ID: <19539@srcsip.UUCP>
Date: 29 Mar 89 16:31:58 GMT
References: <11936@shamash.cdc.com>
Sender: news@src.honeywell.COM
Lines: 43
In-reply-to: tciaccio@hal.cdc.com's message of 21 Mar 89 15:11:37 GMT


>  I'm interested in any products that do diagnosis of electronic devices
>  given a functional model of the device's parts.  
[stuff deleted]
>  Are there any other products or interesting work in this area?  Or
>  should I stick to rule-based implementations for real-world
>  troubleshooting problems?  What's been everyone's experience?
>
>  --- Thomas R. Ciaccio, Control Data Corporation
>		     2800 E. Old Shakopee Road, m/s HQM234
>		     Bloomington, MN.  55425 EMail address -
>  tciaccio@shamash.cdc.com
>

We've been working about four years on a large model-based system for the
Air Force called Flight Control Maintenance Diagnostic System.  It's for
F-16 flight controls, a quad-redundant fly-by-wire system.  Our model
covers the entire flight control system, 43 line replaceable units
(LRUs) containing over 800 sub-lrus (not to mention the signals, panels,
cables, ....)  Total is about 6000 frames.

We started working in KEE, but as the model grew, we've moved out of that
environment for everything but authoring the model.  It's in field test now
at McDill AFB.  It runs on a Compaq 386/20 PC compatible, and we're moving
to host it on a laptop.

We've been pretty happy (bias?  Ha!) with the model-based approach, since
it's easy to build tools to check the validity of the model, something that
rule-based approaches have a notoriously tough time doing (and something
that is of paramount importance to the AF).  We've used the same approach
for factory testing of production line products such as torpedos, and are
currently using it for commercial avionics and on Space Station.

It's also credible to say that this model-based system will run in
real-time, (see AI Magazine, Spring '88 for a discussion of real-time
rule-based systems and their roadblocks), since it has a guaranteed
response (cycle) time, after which it has an answer which incrementally
improves with more cycles (more information).

Harry A. Funk                             Voice: (612)-782-7396
Honeywell Systems and Research Center     Inet:  funk@src.honeywell.com
3660 Technology Dr.   MS:MN65-2400        UUCP:  funk@srcsip
Minneapolis, MN 55418 			  Bang:  {umn-cs,ems,mmm}!srcsip!funk