Change-Point Detection of Natural Frequency Using Dynamic Model Selection and Clustering

Abstract

In order to apply structural monitoring into practice, it is indispensable to develop a method for change-point detection of bridge vibrational properties. In this study, the authors have developed a methodology using sequential prediction, dynamic model selection, and clustering, with the purpose of evaluating vibrational properties and its change timings. In order to examine the validity, the time series that had the change in frequency was calculated, and proposal method was applied. As a results, it was found that the natural frequency is detected. As for change timings, these were evaluated as change intervals. On the other hand, when the proposed method was applied to the acceleration response of a bridge when a train passes, it was found that bridge frequency can be extracted stably, and it is difficult to detect change timings when the change is small. Based on these results, factors were studied, and some solutions to the problems were proposed.