Reassessment of post-construction residual settlement of a bridge approach embankment using Bayesian back analysis

As part of a highway upgrade project in northern New South Wales, Australia, a bridge was constructed over deep soft soils improved by preloading and prefabricated vertical drains (PVDs). Shortly after the bridge opened to traffic, the bridge approach slab settled beyond the serviceability limit. Although slab jacking was implemented, subsequent monitoring revealed settlement again exceeded the predicted upper bound, prompting a reassessment of long-term residual settlement and mitigation strategies. However, this reassessment is challenged by discontinuous monitoring data, instrumentation changes and uncertainty in settlement offsets. Early settlement measurements (May 2017 to February 2018) were taken away from the final embankment location due to a design-stage realignment that shifted the southern abutment. Monitoring was halted during abutment construction and resumed from July 2019. To overcome these challenges, a Bayesian back analysis framework was adopted to calibrate both dataset offsets and soil parameters. The analysis showed that using only the post-construction monitoring data provides the closest fit to the measurements and a reliable prediction of the ongoing settlement growth. The predicted residual settlement over the service life ranges from 306 to 444 mm, with an average value of 385 mm. Sensitivity analyses indicate that slight variations in fill unit weight, due to heavy compaction, and in soft soil thickness, influenced by bridge realignment, have limited impact on settlement predictions due to compensating effects within the Bayesian model. This study also demonstrates the value of probabilistic approaches for assessing long-term settlement under data discontinuities and soil uncertainty, providing insights for similar infrastructure projects.