Probabilistic prediction of long-term deflection in prestressed concrete bridges considering uncertainties in environmental effects, prestress relaxation and traffic-induced cyclic creep

Excessive long-term deflection poses a major threat to the serviceability and safety of prestressed concrete bridges. Although previous studies have applied deterministic and Bayesian inference (BI)-based probabilistic methods to improve long-term deflection prediction, most rely on simplified forward models (e.g., B3 or CEB-FIP) that neglect site-specific environmental conditions, traffic loads, and prestress relaxation—factors that significantly affect deflection evolution. Consequently, existing approaches may yield biased estimations and unreliable predictions under coupled multi-factor influences. To overcome these limitations, this study proposes a comprehensive probabilistic inference framework that integrates a physics-informed forward model with Kriging surrogate modeling and Markov Chain Monte Carlo sampling. The forward model explicitly incorporates the combined effects of environmental hygrothermal variations, traffic-induced cyclic loading, and prestress relaxation, enabling dynamic representation of concrete shrinkage and creep under realistic service conditions. By assimilating observed deflection data, the framework achieves simultaneous updating of multiple deterioration parameters, effectively mitigating modeling uncertainty and improving predictive reliability. Compared with existing BI-based studies limited to material uncertainties, the proposed framework establishes a multi-factor coupling and data-driven updating mechanism that enhances the interpretability and robustness of long-term deflection prediction. This work provides a new basis for reliability assessment and maintenance decision-making of PC bridges under complex service status.