Diagnosing multiple faults in intelligent controls and automated systems

Abstract

The discipline of mathematical statistics has developed techniques of handling multiple events or what are called distributions of multivariate nature. An attempt is made to show how these traditional techniques can be applied to the problem of multiple failures in an expert system context. Generally, in a expert system, a set of hypotheses is proposed on the basis of the symptoms and, through a backward chaining or forward chaining technique, the set of causes is determined for the symptoms. It is in this process that the use of multivariate statistical techniques can be useful. One of the techniques, called principal components analysis, in which a set of symptoms and the covariance matrix of causes can be analyzed, is shown as an example. This yields a set of principal components that can be used to represent a large number of possible values of symptoms in a diagnostic application. An application to intelligent controls is discussed.<>