ABSTRACT
The process of damage identification in bridge structures traditionally involves the extraction of one or more of the numerous modal-based Damage Sensitive Features (DSFs) from vibration data obtained by direct sensor measurement. However the performance of these modal-based DSFs can suffer under changing environmental and operational conditions and are generally limited by their methodology, which assumes linear structural behaviour and signal stationarity. The present paper presents a detailed overview of the development of alternative DSFs derived from vibration characteristics, focusing on their conception, damage sensitivity evaluation and performance robustness assessment. Initially, selected vibration parameters are outlined to the reader before their damage sensitivity is determined on progressive damage test data obtained from a post-tensioned three-span bridge under ambient conditions using supervised machine learning techniques in conjunction with the Minimum Covariance Determinate (MCD) estimator to mitigate uncertainty surrounding sources of excitation. Secondly, the performance robustness of the vibration-based DSFs is assessed on highly non-stationary data obtained from a progressively damaged steel truss bridge subjected to vehicle excitation. Finally, a comparative evaluation of the vibration-based DSFs is made against modal-based DSFs performance from the literature.
