An artificial neural network for eddy current testing of austenitic stainless steel welds

Abstract

An artificial neural network is presented for on-line eddy current testing of austenitic stainless steel welds. Time-domain parameters that are functions of digitized in-phase and quadrature components of probe impedance are used as input to the neural network and the network output, in depth units, is evaluated and displayed continuously. The neural network is trained to recognize disturbing variables such as variations in weld microstructure, lift-off and edge-effect as well as notches of different depth. The neural network is able to detect and characterize longitudinal and transverse surface-breaking notches, despite the presence of disturbing variables.