Applications of improved linear chirplet time frequency representation to Machine bearing fault analysis

Time-frequency analysis is a widely used detection tool for the rotating machinery fault signals. The mechanical bearing fault vibration signals are a non-stationary and non-Gaussian process, it is confirmed that they are α stable distribution, and the characteristic index1<α<2, even the noises belong to α stable distribution process in special circumstances. The existing linear chirplet transform(LCT) methods degenerate, even fail in α stable distribution noise. This paper proposes several robust time–frequency analysis methods to overcome the influence of the impulse noise including. fractional lower order continuous wavelet transform (FLOCWT), fractional low order linear chirplet transform(FLOLCT), fractional low order general linear chirplet transform(FLOGLCT) and fractional low order velocity synchronous linear chirplet transform(FLOVSLCT). The improved methods is applied to compare with the existing methods based on second order statistics in Gaussian and α distribution environments, the simulation results demonstrate performance advantages of the new methods. The features, deficiencies and application scenarios of the improved time–frequency methods have been summarized. Finally, the new methods are applied to analyze the bearing outer race DE fault data contaminated by α stable distribution noise and extract their fault signature, the results illustrate their performances.