Effect of temperature variation on Bayesian anomaly detection in model bridge experiments

Abstract

Vibration-based structural health monitoring (SHM) focusing on changes in vibration characteristics has been studied to improve the efficiency of bridge maintenance and management. Particularly in long-term SHM, which tracks changes in vibration characteristics over a long period of time, attracts attention as a practical bridge SHM method. However, in the long-term bridge SHM seasonal, changes in temperature, traffic, etc. will affect vibration characteristics of bridges. Therefore, it is worth investigating the influence of temperature changes to damage detection of bridges in terms of a long-term bridge SHM. This study investigates influences of temperature changes to vibration characteristics of bridge through a laboratory moving vehicle experiment on a model bridge with artificial damage. As a novel damage sensitive feature (DSF) instead of conventional modal parameters such as frequency, mode shapes and damping ratios, posterior distribution of the system matrix of a multi-dimensional state space model is investigated. The reason for using the novel DSF is to reduce burdens to decide the sensitive vibration mode against damage. The posterior probability distribution of the system matrix is then identified by means of Bayesian inference, and is used for estimating Bayes factor (BF) in terms of a Bayesian anomaly detection. Observations demonstrated that the influence of temperature changes to BF was weaker than that of artificial damage. Moreover, the effect of temperature on BF was relatively smaller than its effect on frequency because the reference feature for BF was extracted from data set of healthy condition with different temperature.