Reconstruction of structural long-term acceleration response based on BiLSTM networks

Abstract

Reconstructing lost dynamic responses is significant for structural condition assessment in structural health monitoring (SHM). Current advanced methods usually employ deep learning models based on convolutional neural networks (CNN) to reconstruct the acceleration response of structures. However, such methods cannot fully mine the temporal correlations among the monitoring data, leading to a relatively low accuracy. Therefore, this study proposes a structural acceleration response reconstruction method based on bidirectional long short-term memory (BiLSTM) networks, which can better mine complex spatial–temporal correlations among the sensor data. A numerical study of a long-span cable-stayed bridge and a practical monitoring data study on the Z24 bridge were conducted to verify the accuracy and efficacy of the proposed reconstruction method. The results show that structural long-term successive acceleration data can be reconstructed with high accuracy using the proposed method despite limited sensors due to the fact that the spatial–temporal correlations of acceleration data can be well extracted. The proposed method outperforms current advanced machine learning-based methods and exhibits exciting application prospects in practical structural acceleration data recovery and damage identification.