Quantitative Ultrasonic Evaluation of Surface-Breaking Crack Height Based on Neural Network

Abstract

A three-layer structured Back Propagation (BP) neural network has been proposed to automatically predict the height of surface-breaking crack, 28 group data including different K values of transducers and the beam path distances obtained directly from the tip echo experiment have been taken into consideration as the training samples. It is shown that the predictions of current network are in good consistence with the real crack heights, the robustness testing demonstrates that the current neural network has its feasible expansibility to be utilized in practical engineering applications.