Test verification of damage identification method based on statistical properties of structural dynamic displacement

Abstract

In this paper, test verification for a method of damage identification using artificial neural network (ANN) based on statistical properties of structural dynamic displacements was carried out. Six fixed beam models are fabricated, and damages on them are simulated by cutting a crack at their bottoms. Some damage samples were used to train the neural network and other samples were used to check the well-trained neural network. During the damage identification using back-propagation neural network (BPNN), the changes of variances (covariance) of structural displacements were adopted as input and location and extent of damage were as the output of BPNN. From the results of test verification in both single-damage case and multi-damage case, it is found that the ANN with the statistical property as damage index can accurately detect the damage location and can identify the damage extent with high precision, which denotes that the proposed method is effective and reliable for damage identification in civil engineering.