Probabilistic prediction of structural response of cable bridges based on structural health monitoring data

Abstract

Recently, structural health monitoring techniques have been widely applied to assess and monitor of the structural condition of a cable bridge. For this purpose, various sensors are often deployed to provide various information such as displacement, strain, tension force, and temperature, and previous studies attempted to predict the structural response of a cable bridge based on the collected measurement data. However, it is still a challenging task to predict the structural response and assess the structural condition of a cable bridge because it involves various uncertainties. In addition, since the number of sensors attached to a cable bridge is generally large, it is computationally-expensive and impractical to use all of the sensor data for the response prediction. This study proposes a new method to probabilistically predict the structural response of a cable bridge based on structural health monitoring data obtained from various sensors. To select only meaningful sensor data for response prediction, a new index based on their correlation coefficients is developed. Then, the proposed method employs Gaussian process regression (GPR), a nonparametric Bayesian method, to build a probabilistic prediction model based on the selected measurement data. In this manner, the probabilistic interval (e.g., 95% confidence interval) as well as the mean prediction of the target response can be obtained. To test the proposed method, it is applied to predict the cable tension forces of an existing cable-stayed bridge in the Republic of Korea. The prediction is made based on various types of actual measurement data including cable tension forces, temperature, and wind speeds, and at first, there were a relatively large number of sensors. However, the new correlation-coefficient-based index allows us to rationally select only the sensor data related to the tension of the target cable, which significantly increases the computational efficiency of the prediction. The prediction results show a good agreement with the actual measurement results, and it is expected that the proposed method can be used for the condition monitoring and anomaly detection of a cable bridge.