Bayesian approaches for evaluating wind-resistant performance of long-span bridges using structural health monitoring data

Reliable estimation of wind-induced displacement responses of long-span bridges is critical to evaluating their wind-resistant performance. In this study, two Bayesian approaches, Bayesian generalized linear model (BGLM) and sparse Bayesian learning (SBL), are proposed for characterizing the wind-induced lateral displacement responses of long-span bridges with structural health monitoring (SHM) data. They are fully model-free data-driven approaches, preferable for reckoning the wind-induced total displacement intended for wind-resistant performance assessment. With the measured displacement responses and wind speeds, a BGLM is developed to characterize the nonlinear relationship between the total displacement response and wind speed, where the Bayesian model class selection (BMCS) criterion is incorporated to determine the optimal model. In the model formulation by SBL, both wind speed and wind direction are treated as explanatory variables to elicit a probabilistic model with sparse structure. The SBL cleverly makes the resulting model to exempt from overfitting and generalizes well on unseen data. The two formulated models are then utilized to forecast the wind-induced displacement responses in extreme typhoon events beyond the monitoring scope, and the predicted displacement responses are contrasted to the finite element analysis results and the design maximum allowable displacement under the serviceability limit state (SLS). The proposed methods are demonstrated using the monitoring data acquired by GPS sensors and anemometers instrumented on a long-span suspension bridge. The results show that the SBL model is superior to the BGLM for wind-induced displacement response prediction and is amenable to SHM-based evaluation of wind-resistant performance under extreme typhoon conditions.