Neural network classifier of cracks in steel tubes : Gavarini, H.; Perazzo, R.P.J.; Reich, S.L.; Altschuler, E.; Pignotti, A. Insight, Vol. 38, No. 2, pp. 108–111 (Feb. 1996)

Abstract

In this paper the possibility of using simple neural networks to classify the severity of defects in the inspection of steel pipes by the magnetic flux leakage technique is analysed. A numerical model simulates the field input to the network, and a Monte Carlo approach is used to generate a population of 1000 flaws by varying the parameters that characterise the tube, the defect and the detection process. 10% of these flaws are used to train a neural network comprising two moduli : a first one that performs a principal component analysis of the field, and a second one that is used to assess the crack depth. The trained network is then shown to be able to reduce substantially the number of false alarms generated in the simulated inspection process.