Rolling Bearing Performance Degradation Assessment with Adaptive Sensitive Feature Selection and Multi-Strategy...

Area of Science:

• Mechanical Engineering
• Condition Monitoring
• Reliability Engineering

Background:

• Single vibration features offer limited insight into rolling bearing degradation.
• High-dimensional feature sets can contain redundant information, impacting reliability assessments.
• Signal outliers and fluctuations complicate accurate degradation performance evaluation.

Purpose of the Study:

• To develop an adaptive sensitive feature selection method for rolling bearing degradation assessment.
• To enhance the robustness of degradation assessment models against signal noise and outliers.
• To accurately determine the onset of rolling bearing degradation using a novel approach.

Main Methods:

• Extraction of high-dimensional vibration signal features from rolling bearings.
• Adaptive sensitive feature selection using Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) and K-medoids.
• Multi-strategy optimization of Support Vector Data Description (SVDD) with Autocorrelation Kernel Regression (AAKR) and multi-kernel functions.

Main Results:

• Successfully constructed a sensitive feature set for degradation assessment.
• Improved the SVDD model's resilience to outliers and signal fluctuations.
• The developed model accurately identified the early degradation starting point of rolling bearings.

Conclusions:

• The proposed method effectively enhances the stability and accuracy of rolling bearing degradation assessment.
• Adaptive feature selection and optimized SVDD provide a reliable tool for condition monitoring.
• This approach offers a significant advancement in predicting and managing rolling bearing health.