Scholarlink

This paper proposes a method based on unsupervised learning for anomaly detection in bridges under traveling loads. The subject of the investigation is the San Paolo viaduct, a multi-span bridge located in Catania, Southern Italy, subjected to dynamic monitoring of one span. The bridge was built about thirty years ago. Therefore, it can be assumed that rheological effects do not affect the measurements in the present monitoring. The Neutral Axis (NA) position of the strain increments due to vehicle passages is chosen as the feature to be investigated in two sections. Thus, strain peaks under vehicle passages were firstly extracted and associated NA positions derived, establishing the Probability Density Function (PDF) for each registration day, univariate for each of the two considered sections, and bi-variate considering the two NA positions simultaneously. Finally, the procedure for outlier detection is assessed using the Mahalanobis distance.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Unsupervised Learning Approach for the Analysis of Long-Term Monitoring Data in a Prestressed Concrete Bridge

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Fernando Saitta,
  • Giacomo Buffarini,
  • Alberto Tofani,
  • Francesca Ciarallo,
  • Diego Esposito

摘要

This paper proposes a method based on unsupervised learning for anomaly detection in bridges under traveling loads. The subject of the investigation is the San Paolo viaduct, a multi-span bridge located in Catania, Southern Italy, subjected to dynamic monitoring of one span. The bridge was built about thirty years ago. Therefore, it can be assumed that rheological effects do not affect the measurements in the present monitoring. The Neutral Axis (NA) position of the strain increments due to vehicle passages is chosen as the feature to be investigated in two sections. Thus, strain peaks under vehicle passages were firstly extracted and associated NA positions derived, establishing the Probability Density Function (PDF) for each registration day, univariate for each of the two considered sections, and bi-variate considering the two NA positions simultaneously. Finally, the procedure for outlier detection is assessed using the Mahalanobis distance.