Evaluating pod-based unsupervised damage identification using controlled damage propagation of out-of-service bridges

Abstract

The effectiveness of Proper Orthogonal Decomposition (POD) for damage identification on out of service, rural bridges is examined using field data. Three similar, out of service, simply supported bridges, consisting of steel beams supporting a concrete deck, were tested prior to demolition and replacement. Strain time histories were recorded using optimal sensor networks with a vehicle of known weight crossed each bridge in different transverse locations and directions, and at varying speeds. The effectiveness with which the developed POD based methodology for damage identification was evaluated by progressively flame cutting select steel beam flanges and webs on each bridge. Snapshot matrices of structural response as measured by the sensors were used to create Proper Orthogonal Modes (POMs), and were obtained for each damage level and test, with the POMs being used to identify if changes in structural response were detected. The accuracy with which the POMs identified damage locations and their sensitivity to vehicle location and direction were evaluated. In addition, a previously developed novelty detection framework proposed by some co-authors was applied to identify changes after damage. It is shown that POD and the developed novelty index can effectively identify certain levels of imposed damage in tested bridges under known loads.