2018 Volume 67 Issue 2 Pages 143-150
This study proposes an anomaly detection method for bridges using Bayesian inference, aiming at efficient inspection based on vibration monitoring. In the proposed method, firstly a posterior distribution of the parameters composing multivariate auto-regressive model is acquired from a bridge under healthy condition by means of Bayesian inference. Secondly, based on the posterior distribution representing vibration of the healthy bridge, a Bayes factor is calculated to detect change in the modal properties caused by damage. To investigate feasibility of the proposed method for damage detection, this study utilized data from a field experiment on an actual steel truss bridge whose truss member was artificially severed. The proposed method detected two different damage levels successfully. A damage indicator previously investigated by the authors is also evaluated with respect to the experimental data, and compared with the proposed method.
