Feature Waveform Extraction Based on Shape- and Position-Adapted Nonparametric Waveform Atoms

In this paper, we introduce a novel method for extracting feature waveforms from signal, which is based on nonparametric waveform atoms. Using a template signal that contains a prior information, a set of basis functions is obtained firstly by means of a uniform filter bank and then a nonparametric atom that is described by a series of discrete data are constructed. The filter bank makes the waveform atom shape-adapted and a delay expansion of the subbands of the filter bank makes the atom position-adapted. Using the constructed atoms, an algorithm for extracting waveforms from signal can be developed based on singular value decomposition (SVD) and matching pursuit (MP). Examples from the simulation analysis and the damping experiment on a rotor-bearing system have confirmed the proposed method. It is shown that the constructed waveform atom can adapt itself to the variations of the feature waveform in the observed signal.