Structural Damage Identification in Bridges Using a Stacked Autoencoder Neural Network

Abstract

Bridge structures are affected by various factors such as the natural environment and traffic load for a long time, which may cause structural damage identification (DI), thus affecting their performance and safety. This paper innovatively combines the stacked autoencoder neural network with curvature modal analysis. The DI method based on curvature modal is to use the curvature modal difference as an indicator for DI. In bridge damage identification, a method combining curvature mode and flexibility matrix is proposed, which is fused into autoencoder neural network to realize the function of damage location. In the test, the key features of the data are extracted through L2 regular term, and the method effect is verified by establishing a simply supported beam model through ANSYS. The identification accuracy of this model in bridge DI is as high as over 78%, and its highest can reach 85%, and the average identification accuracy is 82%. The results show that this method can identify specific damaged units and reflect the relative degree of damage, regardless of single damage or multiple damage conditions. Therefore, the bridge DI identification model based on stacked autoencoder neural network can be applied to real-time identification and analysis of bridge structures to help provide reliable bridge monitoring data support.