No abstract

The influence of varying environmental/operational situations on damagesensitive properties is a vital problem when the vibration responses are employed for structural damage detection. In this work, singular spectrum analysis (SSA) and the statistical control chart are combined to develop a new damage identification approach for structural damage detection under variable environmental/operational situations. The SSA is utilized to decompose the measured frequency sequence into the sum of independent components, including a tardily changing trend, oscillatory, and noise components. The trend-related decomposed components can be identified by finding the corresponding slow-varying eigenvectors. The trend components induced by varying environmental/operational conditions are discarded, while the remaining ones, including oscillatory and noise components, are selected to reconstruct the new frequency sequence. Hence, the impact of variable environmental/operational situations on the original frequency sequence is removed. The reconstructed frequency shift from its reference status (undamaged status) is utilized to construct the damage features, which are employed as the control chart samples. Accordingly, the damage features obtained from the undamaged status are utilized to calculate the control limits. The subsequent damage indices in unknown status are monitored considering the control limits. A considerable number of damage indices out of the control limits range show a system transformation from an undamaged status to a damaged one. The feasibility of the presented approach is analytically and experimentally evaluated through an offshore platform in the presence of a white noise excitation.

In this work, an entropy based two-step structural damage identification method under seismic excitation is proposed. The measured signals are decomposed by means of wavelet packet transform, and the wavelet entropies are obtained on the basis of the information entropy theory. In the first step, the damage alarming indices, calculated with the wavelet entropies in undamaged and damaged conditions, are used as samples for Shewhart individuals control chart to alarm the structural damage. In the second step, the damage localization indices, constructed by calculating the differences of curvature of wavelet entropies in undamaged and damaged conditions, are fed to the back-propagation neural network to identify structure damage location. The numerical simulation and shaking table model test of an offshore platform under seismic excitation are implemented to verify the feasibility of the proposed two-step structural damage identification method. The results show that the proposed method needs only the non-stationary output data and has low sensitivity to signal noise.