Inspection of Ferromagnetic Support Structures From Within Alloy 800 Steam Generator Tubes Using Pulsed Eddy Current

Thumbnail Image
Buck, Jeremy
Signal Processing , Steam Generator , Pulsed Eddy Current , Principal Component Analysis , Artificial Neural Network , Nondestructive Testing
Nondestructive testing is a critical aspect of component lifetime management. Nuclear steam generator (SG) tubes are the thinnest barrier between irradiated primary heat transport system and the secondary heat transport system, whose components are not rated for large radiation fields. Conventional eddy current testing (ECT) and ultrasonic testing are currently employed for inspecting SG tubes, with the former doing most inspections due to speed and reliability based on an understanding of how flaws affect coil impedance parameters when conductors are subjected to harmonically induced currents. However, when multiple degradation modes are present simultaneously near ferromagnetic materials, such as tube fretting, support structure corrosion, and magnetite fouling, ECT reliability decreases. Pulsed eddy current (PEC), which induces transient eddy currents via square wave excitation, has been considered in this thesis to simultaneously examine SG tube and support structure conditions. An array probe consisting of a central driver, coaxial with the tube, and an array of 8 sensing coils, was used in this thesis to perform laboratory measurements. The probe was delivered from the inner diameter (ID) of the SG tube, where support hole diameter, tube frets, and 2D off-centering were varied. When considering two variables simultaneously, scores obtained from a modified principal components analysis (MPCA) were sufficient for parameter extraction. In the case of hole ID variation with two dimensional tube off-centering (three parameters), multiple linear regression (MLR) of the MPCA scores provided good estimates of parameters. However, once a fourth variable, outer diameter tube frets, was introduced, MLR proved insufficient. Artificial neural networks (ANNs) were investigated in order to perform pattern recognition on the MPCA scores to simultaneously extract the four measurement parameters from the data. All models throughout this thesis were created and validated using experimental data. The final ANN models could provide estimates to within 2% of hole diameter and 3% of fret depth. Estimates of hole ID and tube position were further improved when considering fret depth as an input, which could occur if fret information was available. ANN models proved robust to measurement error, as would be encountered in real inspection settings.
External DOI