Speaker
Description
Abstract. Temperature fluctuations in cold climates can significantly distort the dynamic characteristics of bridges, leading to misinterpretations in structural health monitoring (SHM) systems. This study proposes a novel unsupervised learning framework designed to robustly mitigate the influence of freezing temperatures on bridge eigenfrequencies, thereby enhancing the reliability of SHM. The approach integrates Gaussian Mixture Modeling (GMM) to probabilistically cluster structural response data into local environmental regimes, combined with Local Principal Component Analysis (LPCA) to reconstruct and normalize eigenfrequency variations within each cluster. By addressing the nonlinear effects of cold climates, the framework effectively isolates and removes environmental biases from modal properties without relying on external temperature measurements. The method is validated using long-term monitoring data from the Z24 Bridge in Switzerland, which was subjected to severe freezing conditions. Results demonstrate that the proposed hybrid learning approach successfully eliminates abrupt frequency shifts during freezing periods and maintains consistent dynamic behavior across varying operational scenarios. The integration of probabilistic clustering and localized dimensionality reduction significantly enhances the robustness of SHM systems under environmental variability. This research represents an important advancement for SHM in cold regions, offering a practical, scalable, and data-driven solution for more accurate structural condition assessment under extreme weather effects.
