Suppressing influence of measurement noise on vibration-based damage detection involving higher-order derivatives

Aimed at minimizing the influence of measurement noise on the precision and robustness of vibration-based damage detection (particularly for those approaches involving higher-order derivatives of vibration signals), three independent noise reduction strategies were proposed, including low-pass wavenumber filtering (LWF), adjustment of measurement density (AMD), and hybrid data fusion. Residing on a spectrum analysis, LWF suppresses the noise influence by filtering the measurement noise out of a specific wavenumber domain whilst remaining damage-associated signal features only. AMD is based on an inherent correlation between the degree of noise influence and the density of measurement points. By selecting an optimal AMD, the effect of measurement noise on detection precision can be alleviated. The hybrid data fusion amalgamates signal features acquired under different measurement conditions, giving prominence to the salient features in common (i.e., those pertaining to damage) whereas suppressing less salient features in individuals (e.g., measurement noise and uncertainties). Three proposed noise reduction techniques were validated respectively in simulation, and then applied experimentally to the detection of damage in a plate-like structure under a noisy measurement condition. Satisfactory noise suppression was achieved, demonstrating the effectiveness of the proposed noise reduction strategies for vibration-based damage detection.