Harnessing data structure for health monitoring and assessment of civil structures: sparse representation and low-rank structure
Doctor of Philosophy
Civil structures are subjected to ambient loads, natural hazards, and man-made extreme events, which can cause deterioration, damage, and even catastrophic failure of structures. Dense networks of sensors embedded in structures, which continuously record structural data, make possible real-time health monitoring and diagnosis of structures. Effectively and efficiently sensing and processing the massive sensor data, potentially from hundreds of channels, is required to identify (update) structural information and detect damage as early as possible to inform immediate decisionmaking. Different from traditional model-based and parametric methods that usually require intensive computation and expert attendance, this thesis explores a new data-driven methodology towards rapid, unsupervised, and automated system identification and damage detection of structures as well as data management by harnessing the data structure itself. Specifically, the sparse representation and low-rank structure inherent but implicit in the multi-channel structural response data are exploited for efficient data sensing, processing, and management in real-time health monitoring and non-destructive assessment of structures. Numerical simulations, laboratory experiments on bench-scale structures, and real-world structures examples, including seismically excited buildings and a super high-rise TV tower, are investigated.