A sensitive spectrum entropy-assisted Bayesian online anomaly inference method for bearing incipient degradation dynamic detection.

Incipient degradation dynamic detection is crucial for preventing serious accidents in the context of rolling bearing online automatic condition monitoring and preventive maintenance. This article presents a novel framework, cyclostationarity-sensitive spectrum fuzzy entropy-assisted Bayesian online anomaly inference (CSFE-BOAI), to address this challenge. A new health index, CSFE, is first defined by performing the fuzzy entropy measure on the extracted cyclostationarity-sensitive spectra to promote incipient-degradation sensitivity and robustness to interferences. Next, the BOAI procedure for detecting anomalies in continuously arriving CSFEs is derived using the robust generalized T-distribution as the underlying predictive distribution. Eventually, the CSFE-BOAI framework is constructed for bearing incipient degradation dynamic detection, which possesses double confirmation of valid anomalies through the Pauta criterion and cyclostationarity-sensitive spectrum. Experimental verifications are performed on two typical bearing degradation data and one healthy-to-incipient defect data. Results show that CSFE-BOAI enables effective and timely incipient degradation alarm and identification of rolling bearings. The comparisons with the eight advanced health indexes and four anomaly detection approaches demonstrate that CSFE-BOAI has the lowest false and missed alarms and therefore has good deployment potential for practical applications.