Application of Hopfield neural network to structural health monitoring

Abstract

Structural health monitoring (SHM) using artificial neural networks has received increasing<br><br>attention due to robustness of neural networks, better performance compared to conventional<br><br>damage detection methods, and influential pattern recognition capability. This article aims to<br><br>introduce Hopfield neural network (HNN), for the first time, to the SHM community. On this<br><br>basis, a novel damage identification method by the HNN is proposed to detect damage and<br><br>estimate damage severity with the aid of measured mode shapes in undamaged and damaged<br><br>conditions. In this method, these vibration characteristics measured from sensors are used as<br><br>initial conditions in the HNN. A key benefit of the HNN is that this novel neural network is<br><br>inherently able to define a threshold value in such a way that any deviation from this value is<br><br>indicative of damage occurrence. The accuracy and performance of the damage detection<br><br>problems by the HNN is experimentally verified by the I40 Bridge. Results show that the<br><br>proposed method is potentially able to detect damage and estimate the damage severity based<br><br>on the outputs of the HNN.