A machine learning approach for classification tasks of ECT signals in steam generator tubes nearby support plate

Abstract

An automatic classification relying on model based machine learning approach is proposed in the context of steam generator tubes inspection by means of eddy current testing. Such tool could realize a first selection of problematic areas and thus potentially considerably reduce the amount of data experts need to analyse. After the generation of databases of signals covering the configurations of interest, a set of classifiers are trained and compared in terms of performance. In order to mitigate the size of datasets and enhance classification performance, a classical dimensionality reduction technique. Results indicate a good potential of such methods for assisting human experts in the task of ECT signals analysis.