Performance Prediction for Steel Bridges Using SHM Data and Bayesian Dynamic Regression Linear Model: A Novel Approach

Abstract

Understanding expected structural behavior enables the early identification of potential structural issues or failure modes, allowing for timely intervention and maintenance. Guided by this premise, this paper proposes the Bayesian dynamic regression linear model (BDRLM) tailored for predicting the real-time performance of cable-stayed bridges in the face of nonstationary sensor data. Drawing from local linear regression techniques, BDRLM integrates probability recurrence, exhibiting heightened sensitivity to structural behavior shifts. This capability fosters real-time behavior prediction and anomaly detection. Embracing a more pragmatic approach, the model treats the sensor measurement error as an unknown factor. This strategy, complemented by Bayesian probability recursion, refines the error's probabilistic distribution parameters, aligning the prediction process more congruently with field practices. Then, based on structural health monitoring (SHM) data of an actual bridge, the extreme stress of the main girder monitoring sections is dynamically predicted, and a dynamic warning threshold based on prediction updates is proposed. Finally, the time-varying reliability indices of the main girder are predicted and estimated. The effectiveness of the proposed method is validated through an actual application and comparisons of several other commonly used methods. This achievement can provide a theoretical basis for bridge early warning and maintenance with prediction requirements.