Distinguishing signals from noise in eddy current inspection of steam generator tubes

The analysis of eddy current data collected from nuclear steam generator tubes is both very important and very labor intensive. To aid the analysis team we have been developing an automatic data analysis package. It is directed towards multispectral data collected with differential bobbin coil probes. The analysis package calibrates the data against a standard tube, and flags all candidate flaws with a bipolar signature similar to the impulse response of the probe. Three stages of analysis are: preprocessing, wavelet analysis, and fuzzy inference. The wavelet transform is the basis for minimizing the number of false positives caused by noise while the fuzzy inference system further prunes the false positives. Results from tests thus far indicate that the analyzer generally detects the flaws found by humans, but has difficulty distinguishing between very weak flaw signals from cracks near tube support plates. Further development is underway creating better procedures for support plate regions.