ABSTRACT
This study proposes a damage indicator (DI) automatically derived from a set of multivariate autoregressive models estimated from ambient vibration of bridges. The DI evaluates a stochastic distance between a set of reference data (data from healthy bridge) and unknown test data. A statistical hypothesis testing based on a probability distribution of the DI was performed for damage detection. A field experiment on a real steel truss bridge whose truss members were artificially severed was conducted so as to investigate efficacy of the proposed DI for damage detection. The experimental result showed the proposed DI enables to detect damage of three different damage patterns clearly. Efficacy of the proposed DI was also examined by comparing to the previously investigated damage sensitive feature using the experimental data of the same bridge. This comparison also showed efficacy of the proposed DI obtained from a multivariate linear system.
