Data-driven vibrational identification for long-span bridges

Vortex-induced vibration (VIV) is one of the most critical problems in the serviceability assessment of long-span bridges, particularly flexible and lightly damped structures. Bridge owners need to monitor the occurrence of VIVs in near real-time for preemptive actions to potential risk. For this purpose, most of the long-span bridges are equipped with a series of sensors, and this structural health monitoring system (SHMS) accumulates large datasets every day. However, these SHMSs practically fail to satisfy their ideal intention due to the lack of knowledge of bridge operators on VIV. In addition to this, as a long-span bridge is subject to various excitation sources, a threshold-based alarm system with wind speed and vibrational amplitude is not feasible for automated VIV classification. Three main VIVs of long-span bridges in South Korea were actually reported by users, not by SHMS. Here, the big-data analysis with machine-learning algorithms enables this automated VIV classification. This study aims to develop the data-driven framework to classify the VIV of the long-span bridges. Artificial Neural Network (ANN) is implemented for developing a data-driven framework for automated VIV classification of the long-span bridges in real-time. Specifically, this study mainly focuses on following goals (1) Generalize the characteristics of VIV from the monitoring data in the bridge (2) Introduce the efficient labeling process for long-term monitoring data (3) Determine adequate features for VIV classification (4) Demonstrate the feasibility of data-driven classification using actual long-term data.