Structural damage localisation using spectral moments

![CDATA[The objective of this paper is to localize damage in a single or multiple state at early stages of development without demanding any previous data records of the structure in its healthy state. The method starts with transforming time series data to frequency domain (at evenly spaced locations along the structure) by employing power spectral density (PSD) and is then followed by generating moments of different orders in all measured locations. Spectral moments retrieve information directly from PSD without any further manipulation and indicate frequency distribution of the spectrum. It was demonstrated that, although, spectral moment is a non-modal characteristic of a process, it is related to modal parameters of response signal; spectral moment at each location is proportional with its corresponding modal vector. Therefore, it is expected due to damage occurrence, spatial curvature of the spectral moment undergoes a deviation at damage location which can be used as a damage sensitive shape signal spectral moment is a superior alternative to extract features rather than modal data since unlike modal data that provide information at specific frequencies, spectral moments are broadband data which contain abundance of information in the whole frequency range. Moreover, spectral moments contain some substantial statistical characteristics of a process and each moment has its own interpretation. In the case of severe damage, there is considerable variation in the shape signal at damage location but once it is less severe there is no pronounced distinctive change at damages spot. To magnify the discontinuity in the spectral moment curvature at damage location, once damage is less severe, wavelet transform was employed. In the regards, distribution of wavelet coefficient of shape signal is utilized to detect any variation at damage location and is introduced as damage index. Highest values of damage index correspond to sensors located close to the damage locations. The method was applied to a clamped-clamped beam subjected to point force random excitation. The robustness of the proposed damage identification method was examined with respect to the damage severity with various damage conditions in both single and multiple states. Moreover, the feasibility of the method was verified in presence of practical uncertainties such as noise. It was demonstrated that the proposed method is particularly attractive for practical applications as it opens an opportunity for online monitoring of the structural integrity without demanding any previous data records of the structure.]]