Near-Real-Time Identification of Seismic Damage Using Unsupervised Deep Neural Network

Prompt identification of structural damage is essential for effective postdisaster responses. To this end, paper proposes a deep neural network (DNN)–based framework to identify seismic based on response data recorded during an earthquake event. The DNN in the proposed constructed by Variational Autoencoder, which one self-supervised DNNs that can construct continuous latent space input learning probabilistic characteristics. trained using flexibility matrices obtained operational modal analysis (OMA) simulated responses target structure under undamaged state. consider load-dependency OMA results, state represented matrix, closest from measured space. each member then estimated difference between two disassembly method. As numerical example, method applied 5-story, 5-bay steel frame analyses are first performed artificial ground motions create train and test datasets. verified with near-real-time simulation El Centro Kobe earthquakes. example demonstrates DNN-based accurately near-real-time.