Application of Neural Networks to Defect Detection in Cantilever Beams with Linearized Damage Behavior

Abstract

This paper shows the feasibility of using a very simple feed-forward backpropagation neural network for fast and accurate estimation of the location and size of a crack in a cantilever beam. The presented network is trained and tested using data generated by a linear, closed-form, one-dimensional theoretical model of the cracked beam. It is shown that the neural network is a very attractive alternative to presently used methods.