Localized damage detection of structures subject to multiple inputs using distance measure of autoregressive model

In this paper, the distance measure of autoregressive (AR) model is seemed as damage indicator. Two distance measures are discussed: one is the cepstral distance, and the other is the Itakura distance. The distance measures of AR model have been successfully applied in image, speech and neurological signal processing applications. This research explores new applications of two distance measures for damage detection in civil engineering. A five-storey building model is used for performance verification. Verification simulations show efficiencies of both distance-based damage indicators when the excitations are mutually uncorrelated. However, the ability of damage indicators to damage localization is deteriorated when the multiple excitations are mutually correlated as there are strong correlations among them. In practice, the excitations acting on civil engineering structures are mutually dependent and correlated, such as wind and traffic loading or inertial forces induced by earthquake. To overcome this difficulty, a pre-whitening filter is applied to cancellation of correlations of excitations before calculating the damage indicators. To examine the proposed methodology, experiment data from a shake table test have been tested. It can be concluded from the results that, by using the pre-whitening filter, the damage identification ability of the proposed damage indicators improves significantly, especially for damage localization. The damage indicator increases monotonically with damage severity, which provides the potential to damage quantification.