The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9730

Title:Maintaining Uncertain Assumptions for Plausible Fault Diagnosis
Date:Sep 1997
Abstract:An Assumption-based Truth Maintenance System (ATMS) assists a problem solver by providing information about the sets of uncertain assumptions under which each proposition holds. The General Diagnostic Engine (GDE) is a system built onto the ATMS for mdel-based multiple fault diagnosis. The truth value attached to assumptions and dependent propositions in the ATMS are binary and therefore lack the subtleties of numeric uncertainty measures. for this reason, the measurement proposition module of GDE utilises a separate calculus, based upon Shannon's information theory. The major disadvantage of this measurement proposer is its high computational complexity which prohibits its usability in certain situations. Although several approaches have already been suggested to integrate a numeric uncertainty handler in the ATMS, little research has been done to find an alternative measurement proposer. The purose of this project is to extend the ATMS withMYCIN's certainty factor theory and utilise this new version in a GDE-style fault diagnosis system. An alternative measurement proposer is devised which uses the numeric uncertainty values provided by this extended ATMS. Whereas certainty factor theory lacks in richness compared to Shannon's full information theory, we show that the integrated approach provides an effective means of diagnosis at a much lower computational cost.

[Search These Pages] [DAI Home Page] [Comment]