PERFORMANCE MONITORING AND DIAGNOSIS IN A PROCESS ENVIRONMENT USING A PROBABILITY-POSSIBILITY APPROACH.

A methodology which accounts for both random and fuzzy aspects of uncertainty introduced in process control is developed. The random, probabilistic component, quantifies the uncertainty of occurence of an event. The fuzzy, possibilistic component, expresses the imprecision in the meaning of an event. The methodology is based on the concepts of the probability of fuzzy event and of information granularity. It aims at providing an expert system with a framework for assessing the performance level of components and systems in a process environment. Fuzzy algebra is used to propagate the uncertainty in the facts and rules of a rule-based system. New information is provided to the system from sensor data, and epistemic knowledge is provided by a human interface. A fuzzified Bayes formula acts as the link, between past and future states and can be used to update the knowledge-base as well as to estimate anticipated performance.