Dynamic Fault Scenarios For Training A Fuzzy C-Means Clustering Fault Classifier

In this paper, the impact of the choice of the dynamic fault scenarios used for training a Fuzzy C-Mean classifier is investigated. The analysis is carried out with respect to the fault scenarios of the digital Instrumentation & Control (I&C) components of the Lead Bismuth Eutectic eXperimental Accelerator Driven System (LBE-XADS). Three different end states can be distinguished, based on the value of the diathermic oil secondary coolant temperature. The fault characteristics are drawn from different distributions, resorting to different sampling strategies. The dynamic fault scenarios thereby generated are used to train different classifiers based on a Fuzzy C-Mean classification algorithm. The comparison of classification performance allows the drawing of some indications on the coverage requirement of the dynamic fault scenarios used for training the classifiers