Preliminary damage localization based on clustering approach

An essential component of the procedure for assessing the safety level of existing buildings is the characterization of damage, which may lead to stiffness degradation, inaccurate predictions of the structural response, and/or grow and propagate to further elements. Nevertheless, damage identification, even in structures of moderate size, generally entails extensive, costly, and time-consuming inspection campaigns. Consequently, the possibility to individuate the damaged area and to confine to this area the on-site investigation can play a decisive role. Within this framework, the present study is devoted to the localization of damage in truss structures, which are widely employed in bridges and in the construction of sports arenas, stands, industrial buildings, and similar facilities. The procedure described here relies on the analysis of the natural frequencies of the structures and their variations induced by the occurrence of damage. The paper investigates the applicability of classification models to individuate the subset of elements where damage is located. To this aim the correlation between the recorded frequency variations and the spatial position of the damage within predefined regions of the structure is also explored by means of a clustering technique. The effectiveness and accuracy of the proposed procedure are evaluated by applying it to pseudo-experimental datasets generated through finite element models representing various damage scenarios.