Probabilistic bridge fatigue evaluation at virtual sensing locations using kernel density estimation

Abstract

Performing bridge fatigue evaluation using field measurements can be difficult given the amount of data needed for effective assessment, access needed to effectively monitor all fatigue prone locations, and associated power requirements and cost. Several studies have been conducted that estimate strain response at unmeasured locations using indirect measurements and subsequently investigate the quality of the estimation using certain metrics [1]. There is little to no research focusing on pragmatically extending these estimation techniques to probabilistic fatigue assessment. This may be because strain estimation has primarily been successful when applied to numerical and laboratory specimens and perceived potential for difficulties associated with applying developed techniques at scale. This study investigates using data-driven, Singular Value Decomposition (SVD), estimated strains at unmeasured locations for probabilistic fatigue assessment of an in-service, railway, bridge. Before performing strain estimations, SVD Proper Orthogonal Modes (POM) variability was reduced using two classification approaches: k-means clustering and root mean square (RMS); and self-organizing maps (SOMs) and POMs. After estimated strains were obtained, reliability analyses using Kernel Density Estimation (KDE) were utilized to perform probabilistic fatigue assessments. Resulting reliability indices computed using estimated strains were compared against reliability indices obtained using measured strains at the same locations. Results showed that reliability indices computed using estimated strains matched closely with indices obtained using measured strains.