Hierarchical Fisher-information-matrix-based Clustering

Abstract

In this paper, a clustering approach, namely Hierarchical Fisher-information-based Clustering (HFC), is proposed for clustering the similar elements in a structure, based on their effect on the modal parameters and dynamic behaviour of the structure. This clustering approach is indispensable and valuable for any damage localization approach, as usually in practice, the SHM systems include very low number of sensors compared to the number of elements of a structure (or degrees of freedom). Therefore, a clustering approach is a great tool in assessing the possible localization resolution. Furthermore, a clustering approach is necessary to be used for one of the powerful tests, i.e. MinMax test, in Statistical Subspace Damage Localization (SSDL) method. The robustness of the SSDL method on localizing damage in real structures was demonstrated in the literature. In here, the HFC clustering approach along with its effects on the damage localization results for a real test structure, the Yellow Frame, will be presented. It will be shown that the HFC approach can robustly cluster the similar elements of a structure compared to a well-known clustering method, i.e. k-means. The results will be shown and compared.