Speaker
Description
Structural Health Monitoring (SHM) has become an essential practice for ensuring the safety and serviceability of critical infrastructure. However, early damage detection remains a major challenge due to the scarcity of monitoring data and the strong influence of environmental and operational variations (EOVs). To address this, an innovative unsupervised hybrid framework based on Spectral Clustering combined with Density Peaks (SC-DP) is proposed. This method is designed to perform anomaly detection by utilizing only limited modal frequency data, making it particularly suitable for short-term monitoring programs where extensive data collection is impractical. In the proposed framework, spectral clustering is employed in the first phase to partition the training data by identifying its intrinsic structure, without the need for prior knowledge of the number of clusters. Subsequently, the density peaks algorithm is applied in the second phase to recognize damage-related anomalies in the testing data, effectively distinguishing structural changes from environmental effects. The key strengths of the SC-DP method include its reduced dependency on hyperparameter tuning, adaptability to small datasets, and robustness against EOVs. The effectiveness of the proposed approach is validated through two real-world case studies: the Z24 Bridge and the Yonghe Bridge. Comparative analyses demonstrate that the SC-DP framework outperforms conventional methods in detecting early-stage damage while maintaining a low false alarm rate. This research provides a promising tool for practical SHM applications, supporting the development of reliable early-warning systems for large-scale civil structures operating under varying environmental conditions.
