Damage detection for cable-stayed bridge structure using hybrid deep learning network and time-series data obtained from fiber optic sensors

Abstract

In Structural Health Monitoring (SHM), the use of time-series data obtained from sensors has garnered interest from the research community worldwide. With the advancement of sensor technologies, this data is becoming increasingly abundant and complex. However, traditional Machine Learning (ML) methods, such as Artificial Neural Networks, are no longer efficient enough to accurately process and diagnose structural damages based on time-dependent data. To address this issue, this study proposes a novel deep learning approach integrating a 1-dimensional Convolutional Neural Network (1DCNN) and Long Short-Term Memory (LSTM) network to enhance the effectiveness of structural damage detection based on time-series data collected from fiber optic sensors. The efficacy of the proposed method is assessed through a dataset from a cable-stayed bridge in a laboratory setting, constructed at the University of Transportation. Accordingly, the results have demonstrated that the proposed method significantly outperforms traditional deep learning approaches, with accuracy rates on the validation and test sets of 77.5% and 74.1%, respectively.