Use of Artificial Neural Networks in Identification of Modulated Pulses Due to Flaws

The analysis of a number of publications proved that artificial neural networks show much promise in identification of signals and images in nondestructive testing. A three-layered neural network with backward propagation has been used in separating noisy signals due to flaws from spurious signals (due to a slant or separation of an eddy-current transducer) in the process of eddy-current testing. Network characteristics at different numbers of neurons in its layers have been investigated. Probabilities of signal identification at different rms noise intensities have been determined.