Signal processing schemes for Eddy Current Testing of steam generator tubes of nuclear power plants

Abstract

► Approximate the drafting component with a piecewise polynomial function and to eliminate the drafting noise through subtraction. ► Signal indication extraction strategy based on the wavelet decomposition. ► Classify the extracted signals based on factors of signal similarity and signal distance in addition with information of the sensor position and original design data. ► Identify the defect indication by means of both a neural network approach and a statistic method. In this paper, signal processing schemes developed for upgrading the Eddy Current Testing system utilized for In-Service Inspection of Steam Generator (SG) tubes of a Fast Breeder Reactor nuclear power plant are presented. Schemes for the wobbling noise recognition, signal indication extraction, defect signal classification, and the separation of mixed signals are proposed, and are applied to both signals measured in a SG mock-up and signals measured with short tube test-pieces in a laboratory environment. The signal indications are extracted based on a wavelet decomposition and threshold value approach in addition with the signal similarity and signal distance information in case of support plate signals. As it is possible for the position of a signal point to be adjusted based on the signature of the support plates, signals of a welding joint or a bending zone can be recognized based on the design or/and manufactory information. The mixed signals of defects and support plate are processed with the similarity analysis strategy and the identification of defect signals is performed by means of both a neural network approach and a statistic method. Satisfactory extraction and classification results are obtained for all the measured signals, which validated the efficiency of the proposed schemes.