Long-span cable-stayed bridges are prone to significant vibrations under strong wind events such as typhoons, which pose a risk to the bridge functioning and the driving safety of passing vehicles. Structural health monitoring (SHM) systems were widely deployed in most long-span bridges, providing long-term measurements of wind characteristics and vibrations of girders and towers. This paper proposes a data-driven approach to predict the vibration amplitudes of girder and towers for early warning based on state-of-the-art machine learning algorithms. A cable-stayed bridge with a main span of 1088 m is taken as the case study. The monitoring data during a strong typhoon Haikui are extracted to establish the database for training the machine learning models. Wind speed, wind direction, and turbulence intensity are selected as input variables, and girder and tower vibrations are considered as the output. Vertical and lateral vibrations are predicted for the girder, while in-plane and out-of-plane vibrations for the tower. Random Forest (RF) is used for vibration prediction and has demonstrated better accuracy than other typical algorithms. An integrated girder vibration indicator is proposed for early warning considering vertical and lateral directions. Gaussian Mixture Model (GMM) is used to approximate the distribution of the vibration indicator with Akaike information criterion (AIC). The early warning is implemented with monitoring data and predicted results. The proposed approach can guide the bridge operator to manage and maintain bridges during typhoon events and avoid bridge damage and traffic accidents due to excessive vibration.
Read full abstract