Abstract: A method for the extraction of AM-FM signal by sparse signal decomposition based on multi-scale chirplet is proposed, and applied to the fault diagnosis of gearbox with large rotating speed variation. According to the characters of the signal to be decomposed, the proposed method selects the atoms used in signal decomposition with corresponding scale adaptively. Owing to the features of multi-scale in the chirplet library and the introduction of the parameters of frequency offset and slope, this method is superior to the old sparse signal decomposition method, which adopts a single scale, and is more applicable to the decomposition of non-stationary signals whose frequency is time-varying. When a gear with large rotating speed variation has faults , a modulation on mesh frequency will emerge in vibration signals, both the modulation frequency and the carrier frequency will change with the rotating speed accordingly. The modulation frequency and the carrier frequency can be obtained by the proposed method and can be used in gearbox fault diagnosis. The proposed method effectively solves the problems in gearbox fault diagnosis with large rotating speed variation which cannot be disposed by the empirical mode decomposition (EMD) and wavelet method. A simulation example and an application example prove the effectiveness of the method.

Key words: Chirplet base function, Fault diagnosis, Gear, Multi-scale, Non-stationary signal, Sparse signal decomposition