Selection of efficient degradation features for rolling element bearing prognosis using Gaussian Process Regression...

Area of Science:

• Mechanical Engineering
• Reliability Engineering
• Signal Processing

Background:

• Rolling element bearings are critical components in manufacturing, requiring effective health monitoring for maintenance.
• Accurate prognosis and remaining useful life estimation are vital for condition-based maintenance strategies.
• Degradation in bearings causes subtle changes in vibration signals, necessitating advanced feature extraction.

Purpose of the Study:

• To model and predict rolling element bearing fault or degradation trends.
• To evaluate the effectiveness of entropy-based vibration features for bearing prognosis.
• To compare different kernel functions of Gaussian Process Regression (GPR) for accuracy.

Main Methods:

• Gaussian Process Regression (GPR) was employed for bearing degradation trend prediction.
• Shannon entropy, permutation entropy, and approximate entropy were used as vibration features.
• A hybrid metric combining monotonicity, robustness, and prognosability was proposed for feature selection.
• GPR with an ARD exponential kernel was utilized for prognosis with a 95% confidence interval.

Main Results:

• Entropy-based features showed better performance in predicting bearing degradation compared to statistical features.
• The GPR model with an ARD exponential kernel effectively predicted degradation trends.
• The proposed hybrid metric aided in selecting efficient bearing degradation trend features.
• Validation was performed using simulated vibration signals and experimental data.

Conclusions:

• Entropy features are highly effective for monitoring rolling element bearing degradation.
• GPR provides a robust framework for bearing prognosis and remaining useful life estimation.
• The methodology offers a reliable approach for condition-based maintenance in industrial machinery.