Singular Value Decomposition Applied to Damage Diagnosis for Ultrasonic Guided Wave Structural Health Monitoring

A structural health monitoring (SHM) system takes frequent monitoring records from permanently installed transducers on structures, and uses the information to identify potential structural degradation and to proactively maintain the structures. Guided wave testing is an attractive technique for structural health monitoring of large structures, because guided waves can propagate long distance and are sensitive to subtle and hidden damage. In guided wave SHM systems, ultrasonic records are often affected by environmental and operational variations, which produce undesired changes and cause false results. Moreover, although continuous monitoring produces sufficient information regarding structural integrity, we lack a data processing tool to extract, store, and utilize the damage-sensitive information to leverage the accuracy and robustness of damage detection and localization. In this dissertation, we develop a data-driven framework based on singular value decomposition that processes guided wave monitoring records to separate damage-related information from effects of environmental and operational variations. The method decomposes sequential monitoring records to reveal trends of different variations, and identifies the singular vector associated with damage development. Combined with delay-and-sum localization method, we can robustly localize the damage using the right singular vectors, which resemble the scatter signal and are robust to environmental and operational variations. The SVD framework is then refined, by adaptively updating the singular vectors with each arriving ultrasonic record , to achieve online damage detection and localization. The SVD damage diagnosis methodology is applied experimentally to detect and localize damage in plate and pipe structures, both in laboratory tests and in field tests.