Data-Driven Fatigue Failure Probability Updating of OSD by Bayesian Backward Propagation

Abstract

This study introduces a data-driven approach for updating the fatigue failure probability of the orthotropic steel deck (OSD) using Bayesian backward propagation. The OSD in steel bridges is considered as a parallel system composed of two critical fatigue-prone components, namely, the rib-to-diaphragm and rib-to-deck joints. A probabilistic model for fatigue reliability is established based on the equivalent structural stress method and limit state function. The system-level fatigue reliability model is then constructed, taking into account the correlations between limit states of individual components through Bayesian network forward propagation. The key advantage of the Bayesian network-based framework is its ability to perform backward propagation, allowing for the updating of failure probabilities for critical components when the system-level failure of the OSD is observed. Consequently, the proposed approach enables the identification of vulnerable components through data-driven fatigue failure probability updating. Finally, the approach is applied to a real instrumented steel bridge to determine the time-dependent fatigue failure probability at both the system and component levels over its service life. The results show that the component-level fatigue failure probability model will underestimate the fatigue life in comparison to the system-level model. Meanwhile, the proposed method could identify vulnerable components by quantifying the fatigue failure probability of in-service steel bridges.