Robust multisynchrosqueezing transform time frequency technologies with application to fault diagnosis
Haibin Wang, Junbo Long, Changshou Deng, Youxue Zhou
Abstract
Time-frequency analysis (TFA) methods serve as effective tools for analyzing stationary signals.Multisynchrosqueezing Transform (MSST) represents a novel post-processing TFA technology designed for pulse-like signals or noisy environments, aiming to enhance the concentration of time–frequency energy.However, in environments characterized by strong impulsive α stable distribution noise, the time–frequency concentration of existing MSST algorithms, a critical performance metric, is significantly compromised, leading to substantial local deviations. To address this limitation, several robust post-processing TFA technologies based on the fractional lower-order statistics theory have been proposed. These include the fractional lower-order local maximum multisynchrosqueezing transform (FLOLMSST), fractional lower-order improved multisynchrosqueezing transform (FLOIMSST), and fractional lower-order time-reassigned multisynchrosqueezing transform (FLOTMSST), with their computational processes detailedly derived. Numerical validation indicates that the proposed robust fractional lower-order MSST methods outperform existing MSST time–frequency techniques in handling α stable distribution environments. They effectively mitigate the interference of strong impulsive noise while maintaining high time–frequency concentration. Experimental analysis on rotating machinery bearing outer race fault signals demonstrates the efficacy of these robust methods, which can clearly reveal fault characteristics even in complex environments.