Sizing and classification of defects in SG tubes of a Nuclear Power Plant from Remote Field ECT signals by using Neural Networks

This paper presents a Neural Network (NN) based scheme for sizing and classification of defects in Steam Generator (SG) tubes of ferromagnetic material from both measured and simulated Remote Field Eddy Current Testing (RFECT) signals. A novel 2D-3D hybrid database approach of edge FEM method is applied for the rapid computation of RFECT signals due to local defects that is necessary for NN training. A feed forward NN is applied for inverse mapping in addition with a Principal Component Analysis (PCA) process. Several feature parameters of RFECT signals are proposed and adopted as the inputs of the NN, while the 3D sizes are parameterized as binary values and taken as the outputs of the NN. By processing both simulated and measured RFECT signals, it is verified that the proposed scheme is efficient.