Non-stationary signal decomposition approach for harmonic responses detection in operational modal analysis

This paper proposes a non-stationary signal decomposition approach to remove the harmonic responses in the operational modal analysis for time-varying structures. Two time–frequency representations, the non-parametric windowed Fourier transform and the power spectral density estimated by the parametric functional series time-dependent autoregressive moving average model, are combined to decompose the raw multimodal responses into individual unimodal components. According to the histogram shape or the excess Kurtosis value of the components, the harmonic components and structural components can be distinguished and the harmonic ones will be removed in the modal identification. To identify the close modes of a time-varying structure, a time–frequency domain decomposition method based on the power spectral density matrix and singular value decomposition technique is also proposed. The numerical and experimental examples finally demonstrate that the proposed approach can remove the harmonic components effectively and can obtain good modal parameters identification results for time-varying structures with close or even repeated modes.