Classifiability analysis of indications by eddy-current testing of nuclear power plant heat-exchanger tube

Abstract

The contribution treats the topic of data transformation for classification of indications in the field of non-destructive defectoscopy by eddy-currents. Four algorithms for transformation of measured signal indications into suitable vectors of features will be presented. The contribution concentrates on the choice of coefficients from single or multiple vectors of features to define new data representation suitable for classification. Important is classification success and minimal dimension of result vector of features.