Evolutionary fuzzy clustering for the Classification of transients in nuclear components

The classification of objects or patterns is an important area of research with practical applications in a variety of fields. In this paper we are interested in the classification of signal transients for the reliable monitoring and timely diagnosing of nuclear components and systems. These represent fundamental tasks for the operation, control and accident management of nuclear power plants. The problem is tackled within a fuzzy clustering approach. The choice of the metrics upon which the clustering is based is critical for obtaining geometric clusters in the features space as close as possible to the real physical classes. In this respect, here the a priori known information regarding the true classes to which the objects belong will be exploited to select, by means of an evolutionary algorithm of literature, an optimal metrics for the clustering. In case the classification thereby obtained were still unsatisfactory, an iterative procedure is used to split the less compact physical classes in further subclasses.