Fuzzy qualitative diagnosis

Abstract

Purely qualitative reasoning methods suffer from two problems. Measured data values must be classified into exactly one qualitative value and qualitative relations between variables represent only direction but not strength of influence. We have previously developed a constraint-based diagnostic system which searches for the "best" assignment of qualitative labels to all variables using heuristic search. Key elements of the reasoning procedure are 1) deriving unknown variable values by qualitative constraint processing, 2) enumerating possible component behaviors, 3) mapping behaviors into behavior modes (some of which imply faults), and 4) focusing search on promising alternatives. In this paper we describe how a fuzzy set-based representation of variable values combined with fuzzy constraint processing admits fuzzy classification of measurements and improves accuracy and focus of the diagnostic process.<>