Incipient bearing fault diagnosis based on optical fiber sensor and feature-informed geometric partition entropy guided informative frequency band extraction

Abstract

Fiber Bragg Grating (FBG) sensors, with advantages of compactness and immunity to electromagnetic interference, offers a transformative solution for overcoming the challenges of bearing fault diagnosis under harsh operating conditions. The FBG dynamic strain signal can be used for detecting early weak bearing fault signal. However, when the FBG dynamic strain signals are processed through second-order differentiation to derive equivalent acceleration signals, the noise is significantly amplified, which poses significant challenges to extract the weak fault feature. To address this issue, this paper proposes a feature-informed geometric partition entropy (FIGPE) guided informative frequency bands (IFBs) extraction strategy for bearing fault diagnosis based on FBG dynamic strain signal. Initially, the cyclostationarity and impulsiveness of the equivalent acceleration signal of FBG strain caused by bearing faults is revealed. After that, a new indicator named as FIGPE with no prior fault information is developed. This indicator comprehensively considers both the cyclostationarity and impulsiveness of the FBG equivalent acceleration signal. Finally, a bandpass filter based on Linear Piecewise Windowing (LPW) is constructed to isolate fault-related frequency components from the signal, where the FIGPE is guided to adaptively determine optimal center frequencies and bandwidth parameters of the filter based on Bayesian optimization. The results show that the proposed method can effectively identify bearing fault features in FBG equivalent acceleration signals without prior knowledge while suppressing noise. Its effectiveness has also been further verified through simulations, experiments, and comparisons with existing methods.