ABSTRACT
In short to medium span bridges, traffic induced vibration enables vibration monitoring of bridges without interrupting the traffic. Changes in modal properties possibly indicates stiffness loss caused by damages on the bridges. In actual bridges, however, the noisy condition caused by the traffic loadings interrupts modal estimation. Also, the seasonal fluctuation of temperature alters modal properties of bridges. To deal with the uncertainty involved in noisy condition, authors previously proposed Bayesian inference method to quantify uncertainty of modal properties. In the proposed method, the uncertainty involved in modal properties are quantified as posterior distribution using unsupervised machine learning. This study investigates the efficacy of the proposed method under the temperature fluctuation. A half-year monitoring data on a plate girder bridge is adopted for the investigation. On the plate girder bridge, 10 uniaxial accelerometers equipped on the lower flange continuously measure acceleration. Firstly, to clarify that the proposed method detects changes in modal properties caused by temperature change, the proposed method is applied to the datasets separately measured from different seasons. Secondly, to verify the feasibility of the proposed uncertainty quantification for the varying temperature, the samples monitored from different seasons are mixed with each other. The posterior distribution of the modal properties are compared to their temperature fluctuation.
