Weighted demodulation-based spectral fuzzy entropy for bearing prognostics and maintenance

Bearings are vital rotating components used in different mechanical systems. Consequently, these systems are closely related to the prognostic maintenance of bearing faults. Spectral entropy (SE), a nonlinear measure, is a useful technique for detecting dynamic changes in vibration signals collected from bearings. However, in real-world applications, extraneous noise-related frequency components often mask spectral characteristics that indicate bearing faults. Consequently, direct application of SE fails to detect the early inception of bearing faults, consistently track fault progression and predict the remaining useful life (RUL) of faulty bearings effectively. To overcome these challenges, this paper proposes a novel measure that enhances the extraction of fault-specific spectral features through weighted demodulation of the analysed vibration signal before computing its spectrum. Subsequently, addressing the limitations of Shannon entropy, fuzzy set theory-based entropy computation is employed to quantify the information content of the computed spectrum. This proposed measure is termed weighted demodulation-based spectral fuzzy entropy (WDSFE). The WDSFE is validated using two run-to-failure experimental bearing datasets. Results demonstrate that WDSFE surpasses conventional SE in early fault detection, consistent fault tracking and RUL prediction. Additionally, WDSFE outperforms other entropy methods such as fuzzy entropy (FE) and cumulative spectrum distribution entropy (CSDE) in bearing prognostics.