Field-data-driven assessment of an in-service skewed integral abutment bridge assisted by deep learning

Abstract

Integral abutment bridges (IABs) exhibit complex and evolving structural behavior due to their interaction with the foundation and surrounding soil. Long-term effects, such as backfill soil ratcheting and cumulative in-plan superstructure rotation, can lead to unexpected structural responses that are not fully understood. This study analyzes more than 3 years of superstructure displacement, abutment rotation (tilt), and temperature data, collected at 0.5 Hz, from an in-service two-span skewed (45°) steel I-girder bridge with integral abutments and staggered-X-cross-frames. The study advances the understanding of IAB behavior, including behavioral anomalies, which could assist in the interpretation of stress deviations reported in previous studies. Findings show that the monitored IAB presents changing behavior, as well as an accumulation of transverse displacement and abutment tilt over time. Analysis indicates that boundary condition representation in finite element models should incorporate the flexibility provided by soil and pile deformation to accurately reflect field behavior. It is further observed that abutment tilt data displays different trends over short and long periods. Discrepancies between these trends underscore the complexity of IAB behavior under varying temperature conditions. Deep learning techniques, particularly long short-term memory models, assisted in identifying these behavioral patterns. This application demonstrates their potential for detecting subtle deviations in bridge response.