Wall Thinning Characterization of Composite Reinforced Steel Tube Using Frequency-Domain PEC Technique and Neural Networks

Abstract

Resin fiber composites reinforcement is used to recover the original mechanical properties of steel tubes subjected to corrosion wall thinning. Pulsed Eddy Current (PEC) technique can perform nondestructive evaluation of this kind of component, due to its capability to penetrate nonmagnetic insulation. Despite the evaluation capability, distinguishing inner surface from outer surface defects is not an easy task for time-domain PEC technique. In this paper, Fast Fourier transform (FFT) in combination with multilayer perceptron (MLP) neural network classifiers are applied to PEC signals and used to detect defects (wall thinning) and also to indicate their position. The tested sample is a carbon steel tube, with 17 mm of composite reinforcement, where two defects were manufactured, one at the inner and another at the outer surface. An automated scanner system is used to obtain C-scan maps, showing the thinning areas. Two feature extraction methods are used to produce the input features for the neural network classifier: the coefficients of the FFT; and the parameters of an exponential curve fitted to the FFT coefficients. The results indicate that the MLP neural network correctly recognized the presence of wall thinning and its location with detection efficiencies of 97.4 and 97.0%, respectively. The PEC technique analysis in frequency-domain associated with a neural network classifier seems to be a promising alternative to identify the position of defects in composite reinforced steel tubes.