Blind modal estimation using smoothed pseudo Wigner-Ville distribution and density peaks clustering

The determination of the number of active modes is always an obstacle to the identification of dynamic systems. To tackle this issue, a new blind modal estimation method is proposed through which the number of active modes does not need to be presupposed, even in underdetermined blind source separation cases. Firstly, the spatial time-frequency distribution matrices are constituted from the smoothed pseudo Wigner-Ville distribution of acceleration signals. The auto-source time-frequency points that contain modal information about only one mode are selected and then the density peaks clustering algorithm is introduced to identify the mode shape vectors. Finally, one can reconstruct the time-frequency distributions of modal coordinates and estimate the natural frequencies and damping ratios. The method is verified by a numerical study and a vibration experiment. The anti-noise study demonstrates that the method performs well.