This paper reports our recent endeavor to develop automated, systematic inversion tools by the novel combination of neural networks and finite element modelling for eddy current flaw characterization in steam generator tubes. Specially, this paper describes I) the construction of databases with abundant flaw signals from 2-D axisymmetric flaws with tip variation using finite element models, 2) the extraction and selection of sensitive features for flaw classification and sizing, and finally 3) the inversion of ECT signals by use of two neural networks for flaw classification and sizing. In addition, this paper also presents the performance of proposed inversion tools for classification and sizing of 2-D axisymmetric flaws 1) having symmetric cross-sections with the variation in tip width, and 2) having non-symmetric cross-sections with the variation in tip deviation.
展开▼