Ambient noise-based passive reconstruction of dispersion curve in a thin-plate with complex boundary interferences using sparse learning

The passive reconstruction of dispersion curve of a plate-like structure using ambient noise has been intensively studied and applied to various structural health monitoring problems. Its principle is to use the cross-correlation between the random guided wave signals measured by two receivers to approximate the impulse response between the two measurement points where the resolution is proportional to the wavelength. This methodology is powerful in many scenarios but problematic for structures with complex boundaries, because boundary reflections contaminate the desired direct propagation components in the cross-correlation such that the estimated impulse response is biased. To address this problem, this paper proposes a sparse learning scheme to decompose the cross-correlation and then to restore the robustness of passive impulse response estimation. First, an overcomplete dictionary considering the features of the cross-correlation components is proposed. We then perform the sparse decomposition on the cross-correlation, from which individual wave packets corresponding to different propagation paths are obtained and this enables a precise dispersion curve reconstruction. The efficiency of the proposed method is demonstrated via a passive experiment where the ambient noise is excited by spraying air-jet onto the surface of a plate. Additionally, the method is successfully applied to the passive detection of ice accretion on plates, which has the potential for passive online monitoring of aircraft structure icing.