Seismic damage identification of cable-stayed bridge in near-real-time using unsupervised deep neural network

Abstract

Prompt damage identification of infrastructure systems is essential for effective post-disaster responses. However, most infrastructure systems have a high level of structural complexity, making damage identification extremely difficult. To overcome the challenge, the authors recently proposed a deep neural network (DNN) based framework for identifying the seismic damage based on the structural response data recorded during an earthquake event (Kim and Song, 2022). The DNN of the proposed framework is constructed by a Variational Autoencoder, one of the self-supervised DNNs capable of constructing a continuous latent space of input data by learning probabilistic characteristics. The DNN model is trained using the covariance matrices of the snapshot of the response data obtained from the undamaged structure. To consider the load-de-pendency, the undamaged state of the structure is represented by the covariance matrix, which is closest to that obtained from the measured seismic response in the latent space. To identify the severity of the structural damage, a structural damage index based on the difference in the covariance matrices is introduced. This paper improves the DNN-based framework to facilitate its applications to complex structural systems such as the Incheon Grand Bridge, a reinforced concrete cable-stayed bridge in South Korea. To generate train, validation, and test datasets, structural analyses are first performed under the ground motions from the PEER-NGA strong motion data-base. The proposed framework is verified with near-real-time simulations using ground motions with various time steps from the test dataset. The example shows that the proposed framework can accurately identify seismic damage of the complex structural system in near-real-time.