Merged LSTM-based pattern recognition of structural behavior of cable-supported bridges

Abstract

Structural responses of bridges occur based on their structural characteristics and conditions. After the structural pattern is identified from the long-term measured response datasets, the structural responses can be evaluated and predicted using a pattern model. In the absence of significant variations in the structural condition, the difference between the predicted and measured responses is negligible. Otherwise, the differences can be identified, and this would be evidence of the variation in the structural condition. Therefore, the structural pattern model can be used effectively to investigate variations in the structural state and conditions. This study proposes an effective structural pattern recognition method using deep learning. A merged model is proposed by combining deep neural network (DNN) and long short-term memory (LSTM) algorithms to handle long-term responses from various sensors in the time domain and reflect statistical properties. Long-term (five-year) measured response datasets of an existing cable-supported bridge were used to validate the proposed method. According to the study, the proposed method can effectively identify the structural behavioral pattern of a cable-supported bridge.