Robust Incipient Fault Diagnosis of Rolling Element Bearings Under Small-Sample Conditions Using Refined Multiscale...

Area of Science:

• Mechanical Engineering
• Data Science
• Artificial Intelligence

Background:

• Aero-engine reliability hinges on bearing health, but early fault detection is difficult with limited data.
• Existing multiscale entropy methods lose information and are unstable with small datasets.

Purpose of the Study:

• Develop a robust diagnostic framework for bearing fault diagnosis under small-sample conditions.
• Ensure feature consistency and classification stability with minimal training data.

Main Methods:

• Proposed an integrated approach combining Refined Time-shifted Multiscale Rating Entropy (RTSMRaE) and Animated Oat Optimization (AOO)-optimized Extreme Learning Machine (ELM).
• RTSMRaE uses a refined time-shift and dual-weight fusion to preserve features and reduce noise.
• AOO optimizes ELM weights and biases for improved stability and generalization.

Main Results:

• Achieved 99.47% diagnostic accuracy with a standard deviation of ±0.48% using only five training samples per class.
• Demonstrated superior diagnostic robustness and computational efficiency on laboratory and real-world aviation bearing data.

Conclusions:

• The RTSMRaE-AOO-ELM framework provides a promising solution for intelligent condition monitoring in data-scarce industrial environments.
• The method offers enhanced reliability and stability for aero-engine bearing fault diagnosis.