MVILAIN@G.BBN.COM (Marc Vilain) (03/04/88)
BBN Science Development Program
AI Seminar Series Lecture
REPRESENTATIONS FOR MODEL BASED TROUBLESHOOTING
Walter C. hamscher
MIT AI Lab
(hamscher@ht.ai.mit.edu)
BBN Labs
10 Moulton Street
2nd floor large conference room
10:30 am, Tuesday March 15
Model based troubleshooting is fundamentally about modeling. Its goal
is to apply a general troubleshooting engine to a new domain by
providing only a new domain model, so it is essential to know not only
what relation the model should bear to the real physical device being
diagnosed, but also what features the resulting model should include by
virtue of its intended use in troubleshooting. Since every model
embodies some abstractions, this is just another way of saying that it's
essential to know the useful abstractions for the task at hand.
This talk presents a methodology for model based troubleshooting of
board-scale digital circuits that emphasizes the importance of
appropriate temporal abstractions for coping with behavioral complexity.
The result is a remarkably coarse representation for digital circuit
behavior that often yields as much diagnostic resolution as traditional
circuit models, in spite of its simplicity. In the same spirit, the
importance of appropriate representation of circuit organization is
emphasized, and the result is a primary representation of the physical
organization of the circuit, along with a more familiar representation
of functional organization.
-------