Artificial Neural Network for On-Line Eddy Current Testing

A new methodology has been proposed to carry on-line multifrequency eddy current testing. This uses a feed forward and error back propagation ANN in conjunction with on-line data acquistion. This methodology digitizes the in-phase and quadrature components of eddy current signals and uses them directly as the input vectors of the network and displays the network output, on-line. The proposed methodology has been successfully applied for defect detection and evaluation in small diameter stainless steel tubes with cold pilger noise and in austentic stainless steel welds with surface roughness, o ferrite and conductivity variation and also for oxide scale thickness measurement in heat exchanger tubes made of copper. As the output of the proposed methodology is directly available in the user defined units, with suitable thresholds, the methodology holds promise for shop-flor applications.