Wavelets transforms and fuzzy logic in the eddy-current inspection of nuclear power plants steam generator tubes

Nuclear power plants steam generators around the world have presented early damage history in their tubes, caused either by design errors or by inappropriate operation, which besides reducing the availability and the safety of the nuclear power plants it also generates heavy economical burden. To monitor the steam generators operational condition, the Eddy Current testing of their tubes is the non destructive method used to detect, localize, classify and to size the defects. The inspection is performed by inserting probes with coils in the tubes generating a signal correlated to the defect. These signals produced by the probe electric circuit are composed by the resistance and the inductive components which can be combined to produce a Lissajous figure in the complex plane. However, Eddy-Current signals contain noise which induce subjectivity inducing to errors in the inspector diagnosis. It is not uncommon to have different diagnosis from two inspectors about the same signal. The present work has the objective of supplying a methodology to analyze the signals which could help the inspector in the difficult task of interpreting the Eddy Current signals. It is proposed a method to remove the noise based on Wavelets Transforms. It is also proposed a normalization in the signal phase angle measurements. Furthermore, two additional characteristics are also studied, namely: the signal amplitudes and the widths of the Lissajous petals. The use of a Fuzzy Logic based inference engine is also developed and its use is demonstrated to be viable. The defects studied in this work are those which produces volumetric changes in the material. In order to test the proposed methodology, several artificial defects were produced in tubes using different types of materials like: brass, 316L stainless steel and Inconel 600 to produce a experimental data base. An Eddy-Current inspection equipment, the MIZ-17ET was used. Around 1000 time series signals of defects were acquired through the LABVIEW program which were read by the Reading and Visualization Signals Module (MLV) of a program developed in MATLAB platform. These signals were pre processed by Wavelets Transforms to remove Data Acquisition Noise, Material Noise and Lift-off effect and the denoised signal shows that the method is appropriate. A Feature Extraction Module (PPS) was also developed under a MATLAB environment to generate the proper characteristics. The signals characteristics obtained by this module are then correlated with defects and used as inputs to a Fuzzy Logic based inference engine to help in the diagnosis. The entire methodology was tested using the Inconel 600 and 316L stainless steel tubes with defects data base. The results show that the proposed methodology is viable. (author)