An Artificial Intelligence Technique to Characterizae Surface-Breaking Cracks

A neural network with an analog output is presented to determine the angle of inclination of a surface-breaking crack from ultrasonic backscattering data. A neural network which was trained by the use of synthetic data set to estimate the depth of a crack, assuming that the inclined crack angle is known, was presented earlier[1,2]. In this study, a neural network estimates the angle of inclination of the surface-breaking crack, assuming that the depth of the crack is 2.0mm, by utilizing the waveforms of backscattered signals from the crack. The plate with a surface-breaking crack is immersed in water and the crack is insonified from the opposite side of the plate. The angle of incidence with the normal to the insonified face of the plate is taken to be 18.9°. The neural network is a feed-forward three layered network. The training algorithm is an error back-propagation algorithm which has been discussed in Refs. [3,4]. The theoretical data obtained by the boundary element method are used for the training. The performance of the trained network is tested by synthetic and experimental data