Bayesian Damage Detection for Bridges under Noisy Condition

Abstract

This study is intended to verify validity of an efficient damage detection method by means of a Bayesian approach especially for noisy operational condition. A Bayesian inference was adopted to the regressive model representing bridge vibration. The posterior distribution for the regressive coefficients provides reasonable damage-sensitive features. Bayesian hypothesis testing is formulated using a Bayes factor, which is defined as a ratio of marginalized likelihoods to detect anomaly in the damage-sensitive features. Feasibility of the proposed method under noisy condition is examined via a field experiment on a continuous Gerber-truss bridge whose truss member was artificially severed. The proposed method robustly detected a damage considered in the experiment even under the varying traffic loadings.