Neuro-fuzzy pattern classification for fault diagnosis in nuclear components

The classification of patterns of measured quantities for diagnostic purposes is an important area of research with practical applications in a variety of fields ranging from industrial to medical. In particular, the classification of faults in nuclear components represents a fundamental task for the operation, control and accident management of nuclear power plants. In this paper, the fault diagnostic problem is tackled within a neuro-fuzzy approach to pattern classification. An important practical issue for the applicability of any diagnostic tool is the transparency of the underlying classification model, to allow for physical interpretation of the relationships between the underlying variables and for direct inspection for validation purposes. In this respect, the proposed neuro-fuzzy approach aims at obtaining not only a high rate of correct classification but also a transparent classification model, i.e., readable and easily interpretable from the physical point of view. For this reason, appropriate coverage and distinguishability constraints on the fuzzy input partitioning interface are introduced. The approach is tested on a diagnostic task concerning faults in the gland seals of the pumps of the primary heat transport system of the CANDU 6 nuclear reactor.