This paper depicts an adaptive diagnostic system based on a fuzzypattern recognition approach. The proposed system is designed to operateon-line and to deal with the following characteristics: on-lineadaptation of classes, detection of slow or abrupt changes andstabilization in a new state, on-line creation of new classes. To meetthese requirements, classes are constructed sequentially with a fuzzyagglomerative clustering procedure. Such a clustering procedure requiresonly one pass through the data, the fuzzy prototypes are created oradapted as new observations are gathered. A prototype is labelled as itis created by using a k-nearest neighbours rule with a distance rejectoption. This labelling rule is well adapted for stationary states andabrupt changes. However, this rule does not operate in case of incipientfaults. To deal with this limitation, we define the concept of temporaryprototype. To decide if this prototype is representative of a stationarystate or a transient one we introduce a progressive hypotheses testbased on the activation rate of the prototype. The results of arobustness study are presented. Finally, the diagnosis system operationis demonstrated on a simulated example
展开▼
