Waveform chain code: a more sensitive feature selection in unsupervised structural damage detection

Abstract

Structural health monitoring is of great significance to the maintenance of long-term used structures, as unexpected damage may lead to disasters and economic loss. A new structural damage detection scheme using waveform chain code and clustering is proposed in this work. The waveform chain code features are extracted from the frequency response functions. Compared with the raw frequency response data, these features show the alterations caused by structural damage more evidently. K-means clustering method is used to distinguish the features of intact and damaged states. Unlike supervised learning methods whose training data are labeled, the unsupervised clustering is performed with unlabeled data. An experimental test on a rectangular Perspex plate is carried out for verification. The results show the good performance of the newly proposed scheme and this might suggest its potential application in the real practice.