Machine Condition Monitoring

Bearings are a key component of rotating machines, and their fault diagnosis is critical for safe operation of rotating machines. Since bearings often work under variable speed conditions and their vibrations contain rich information of health conditions, time–frequency analysis (TFA) of vibration signals has been shown to be an effective way to perform bearing fault diagnosis. However, applications of traditional TFA methods for analyzing vibrations from bearings are often constrained by limited time variability and smearing effects. This article proposes an instantaneous frequency (IF) synchronized-generalized stepwise demodulation transform (IFS-GSDT) method for TFA of nonstationary vibration signals. Demodulators of the proposed IFS-GSDT method are first derived as functions of inclined angles formed by IF lines of windowed signals; thus, IF preestimation is no longer required. A spectral kurtosis-guided strategy is then developed to determine optimal inclined angles. To effectively tackle multicomponent signals, the proposed IFS-GSDT method explores a new linear transforming kernel that synchronizes the demodulators to all signal components, and an iteration procedure can be avoided. The proposed method also allows for the signal to be reconstructed when the window length under analysis is fixed. The effectiveness of the proposed method is validated using simulations and measured vibration data. Comparisons between the proposed method and other popular TFA methods are also conducted to demonstrate the superior characteristics of the proposed method.

Time-frequency (TF) features of nonstationary vibrations are indicative of the health condition of rotating machinery and, are also pivotal in analyzing acoustic signals obtained from processes such as bat echo-location. However, the TF features in these nonstationary vibration and acoustic signals are often submerged by strong background noise. This article proposes using the refined matching linear chirplet transform (RMLCT) to enhance the TF features, where the chirp rates are adaptively determined by spectral kurtosis and only the interesting time-frequency representations (TFRs) are retained. With selected chirp rates, a novel transformation kernel is developed, enabling the proposed method to simultaneously process nonstationary multicomponent signals. Moreover, the angle refinement strategy is proposed to improve the noise-handling capability of the proposed method. The signal reconstruction of the RMLCT is also analyzed, demonstrating that signal components of interest can be accurately reconstructed. Numerical and experimental analyses validate the effectiveness of the proposed RMLCT.