Sensor-Informed Probabilistic Bridge Reliability Analysis: A Review.

Abstract

This review examines the integration of sensor data into probabilistic bridge reliability analysis, covering the workflow from sensor selection and data acquisition through preprocessing, computational Bayesian inference and reliability index <i>&#x3b2;</i> computation to remaining service life estimation and maintenance optimization. Sensor selection is treated as a reliability engineering decision, with sensing technologies matched to the target reliability variable and governing deterioration mechanism. Sensor imperfections inherent to long-term structural health monitoring, including drift, bias, noise, missing data, synchronization errors, and calibration uncertainty, are evaluated through their propagation along the reliability analysis chain. Statistical filtering, thermal compensation, anomaly detection, and missing data imputation are examined as essential preprocessing stages for reliable downstream inference. Computational Bayesian methods are organized into core posterior inference techniques, including Markov Chain Monte Carlo, Sequential Monte Carlo, variational inference, approximate Bayesian computation, and neural simulation-based inference, and supporting techniques, including surrogate-based approaches, deterministic filtering, asymptotic approximations, and hybrid Bayesian reliability frameworks. The synthesis indicates that sensor imperfections can propagate into reliability indices and RSL estimates, that preprocessing is reliability-critical, and that no single inference method is universally suitable for sensor-informed bridge reliability assessment. This propagation is demonstrated by the illustrative scenario, in which RSL effects followed the ordering drift &gt; missing data &gt; bias; however, this ordering is conditional on the assumed parameter values and monitoring conditions and does not represent a general ranking of sensor imperfections. A key gap concerns the sensor-to-reliability pathway, as monitoring observations do not directly determine <i>&#x3b2;</i> but require mapping to limit state variables with consideration of identifiability and spatial representativeness. Integrating these elements with RSL estimation, value-of-information analysis, and lifecycle maintenance optimization supports evidence-based bridge management while highlighting challenges in sensor quality, model adequacy, uncertainty propagation, and computational implementation.