Experimental validation and numerical investigation of virtual strain sensing methods for steel railway bridges

Abstract

Railway bridges are subjected to intermittent excitations induced by gravity loads. The monitoring of stress time histories at fatigue prone areas (i.e., “hot spots”) for steel, heavy haul railway bridges is vital for cost-effective maintenance and, most importantly, for maintaining safety. In this regard, strain measurements are crucial for determining fatigue susceptibility and, possibly, damage to steel bridges under daily, cyclical loading conditions. While, in the limit, sensing the response of a steel bridge under operational conditions is the best way to identify hot spots to prevent damage and, possibly, collapse, it is impractical if not impossible to install strain sensors at every susceptible location. To address this issue, the current study focuses on virtual sensing of strain time histories on steel railway bridges using sparse response measurements. An augmented Kalman filter (AKF) method was adopted for input-state estimation. Since AKF estimates unknown load and response using a physical model, it is crucial to assess the effects of modeling uncertainties on estimation results. In addition to AKF, Modal Expansion (ME) was adopted for extrapolation of measured response to unmeasured locations. In contrast to AKF, which requires a full physical model, ME only relies on vibration modes. A novel application of the Singular Value Decomposition (SVD) method that facilitates data-driven strain prediction by relying on data obtained from field measurements or models was also developed and examined. The study was completed using simulated experiments and strains measured from an in-service, through girder, steel railway bridge. The three methods were compared and strengths and limitations of each were highlighted.