Machine-learning-based prediction of vortex-induced vibration in long-span bridges using limited information

Long-span bridges are susceptible to wind-induced vibration due to their high flexibility, low-frequency dominance, and light damping capacity. Vortex-induced vibrations (VIVs), which usually occur under in-service conditions, can result in discomfort to users and detrimental effects on the fatigue capacity of structural elements; therefore, accurate VIV assessments are essential in ensuring the vibrational serviceability of bridges. Despite the research efforts of data-driven VIV prediction, the robustness and general applicability of the proposed methods remains challenging, in that each method requires different conditions for the datasets in order to develop machine-learning (ML) models. Furthermore, collecting sufficient VIV datasets (anomaly state) from various operational conditions is impractical, time-consuming, and even impossible in some situations compared with non-VIV datasets (normal state). This imbalance in the dataset could degrade the model performance. To address this issue, this paper focuses on developing a general framework for introducing ML algorithms to predict VIVs with a limited amount of information. To properly replicate the practical cases, two different scenarios are assumed along with the amount of VIV data: (1) no VIV data are available, or (2) only a small number of VIV data can be obtained. A variety of ML-assisted methods are introduced for each scenario to predict VIVs in order to demonstrate the versatility of the proposed framework. The effectiveness and applicability of the proposed framework are demonstrated using actual monitoring data. Different methods are prepared to provide further insight into the ML algorithms used for VIV prediction. The proposed framework in this paper is expected to advance our knowledge and understanding of the application of ML algorithms to bridge systems, which are essential in enhancing resilience against wind hazards.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01856280
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Aug 29 2022 9:27AM