A Low Complexity Rolling Bearing Diagnosis Technique Based on Machine Learning and Smart Preprocessing.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearings are critical components in rotating machinery, and their failure can lead to significant downtime and costs.
• Effective condition monitoring and diagnosis systems are essential for predictive maintenance and operational reliability.
• Existing diagnosis systems often struggle with varying operating conditions and diverse acquisition setups.

Purpose of the Study:

• To develop an accurate and robust diagnosis system for rolling bearings.
• To create a system deployable on embedded electronics with low computational demands.
• To design a fault diagnosis system with minimal dependency on machine working conditions, bearing models, and acquisition systems.

Main Methods:

• Simultaneous measurement of vibrations and machine rotation speed.
• Combination of time-domain methods for fault detection and machine learning for fault location.
• Development of a novel preprocessing algorithm for neural network classifiers to enhance robustness.
• Training using emulated data from an ad-hoc test bench.

Main Results:

• Achieved an overall classifier accuracy greater than 98% using emulated data.
• Demonstrated strong generalization capabilities with accuracies over 97% across varied working conditions, setups, and noise levels.
• Validated the system's distinct generalization capability using real-world vibration data from a different scenario.

Conclusions:

• The proposed rolling bearing diagnosis system is highly accurate and robust.
• The system's low complexity and resource demand make it suitable for embedded applications.
• The demonstrated generalization ability allows for broad applicability across various industrial scenarios.