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

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Online condition monitoring and preventive maintenance of rolling bearings are critical for preventing catastrophic failures.
• Early detection of incipient degradation is essential for timely intervention and avoiding serious accidents.

Purpose of the Study:

• To develop a novel framework, CSFE-BOAI, for sensitive and robust detection of incipient degradation in rolling bearings.
• To enhance the reliability of anomaly detection in bearing health monitoring systems.

Main Methods:

• Defining a new health index, cyclostationarity-sensitive spectrum fuzzy entropy (CSFE), by applying fuzzy entropy to cyclostationarity-sensitive spectra.
• Deriving a Bayesian online anomaly inference (BOAI) procedure using a generalized T-distribution for continuous CSFE data.
• Constructing the CSFE-BOAI framework with double anomaly confirmation using the Pauta criterion and cyclostationarity-sensitive spectrum.

Main Results:

• Experimental validation on bearing degradation datasets demonstrated effective and timely incipient degradation alarming and identification.
• The CSFE-BOAI framework achieved the lowest false and missed alarm rates compared to eight advanced health indexes and four anomaly detection approaches.
• The proposed method shows robustness to interferences and enhanced sensitivity to incipient degradation.

Conclusions:

• The CSFE-BOAI framework provides a reliable and accurate solution for incipient degradation dynamic detection in rolling bearings.
• Its superior performance in minimizing false alarms suggests significant potential for practical deployment in industrial applications.
• This approach advances the field of condition monitoring for predictive maintenance of rotating machinery.