Online unsupervised detection of structural changes using train–induced dynamic responses

• Online unsupervised early damage-detection in a long-span bowstring bridge using train-induced responses. The performance of AR and ARX models as feature extractors was evaluated. Influence temperature, noise, train type speed are efficiently removed from the features. A novel approach to define an adaptive confidence boundary is successfully validated. real-time SHM procedure simulated check proposed strategy’s robustness sensitivity. This paper exploits data-driven structural health monitoring (SHM) order propose continuous online for damage detection based on dynamic responses, taking advantage large-magnitude loading enhancing sensitivity small-scale changes. While such large responses induced by trains might create more damage-sensitive information measured response, it also amplifies effects those measurements environment. Thus, one biggest contributions herein methodology that passage while rejecting confounding influences environment way false positive detections mitigated. Furthermore, this research work introduces adaptable decision threshold further improves over time. To ensure assessment, hybrid combination autoregressive exogenous input (ARX) models, principal components analysis (PCA), clustering algorithms sequentially applied data, moving window process. comparison between obtained (AR) conducted, concluded lead increased due their ability capture cross sensors. PCA proved its importance effectiveness removing observable changes variations or temperature without need measure them, methods allowed automatic classification Since not possible introduce bridge, several conditions were with highly reliable digital twin Sado Bridge, tuned experimental data acquired system installed site, test validate efficiency procedure. strategy be robust when detecting comprehensive set scenarios incidence 2%. Moreover, showed smaller levels (earlier life), even consists small stiffness reductions do impair safety imperceptible original signals.