Bayesian updating of model parameters for jacking force prediction using field observations

Abstract

In recent years, there has been an increased resort to box-jacking as a method for constructing underground utilities and infrastructure, as it avoids the on-street disruption in urban areas. An accurate estimation of the total jacking force requirements during a drive is a key design consideration for determining the capacity of box culverts, the placement of interjack stations, and the overall efficacy of the box-jacking project. Nevertheless, predicting the total jacking force is complicated by site geology, lubrication performance, work stoppages, deviations in alignment, and the imperfect calculation model. This study introduces a Bayesian framework that predicts jacking forces by using pipe-jacking parameters updated from field observations. The proposed framework was applied in two box-jacking monitoring projects in Japan, and the forecasts were appraised though comparison to predictions determined using classical optimization technique, namely particle swarm optimization. The results show that predictions of jacking forces based on prior hypotheses resulted in a significant overestimation of the monitored jacking forces for both drives. In contrast, the proposed framework provided excellent forecasts, highlighting the limitations of prescriptive design approaches in capturing complex geotechnical conditions during tunnelling and the importance of robust back-analysis techniques. Furthermore, the results also show that Bayesian updating coupled with a sliding window approach is a very effective option when there are considerable fluctuations in jacking force development.