Diagnosis Methodology Based on Deep Feature Learning for Fault Identification in Metallic, Hybrid and Ceramic...

Area of Science:

• Rotary electrical machinery and electromechanical systems engineering.
• Advanced materials science, focusing on hybrid and ceramic bearings.
• Data-driven fault diagnosis and machine learning applications.

Background:

• Rotary electrical machinery efficiency is increasing due to technological advances.
• High-performance actuators utilize advanced materials like hybrid and ceramic bearings.
• Existing bearing fault diagnosis methods are often limited to metallic bearings, failing to account for varying vibration patterns across different bearing technologies.

Purpose of the Study:

• To propose a novel data-driven diagnosis methodology for identifying bearing faults in electromechanical systems.
• To develop a system capable of diagnosing faults across diverse bearing technologies, including metallic, hybrid, and ceramic bearings.
• To enhance condition-monitoring strategies for rotary machinery with varied bearing types.

Main Methods:

• A deep learning model utilizing stacked autoencoder structures for adaptive feature extraction from multi-domain signals.
• A feature fusion stage to integrate information from different signal domains, improving discrimination.
• A final classification stage using a softmax layer for accurate bearing fault assessment.

Main Results:

• The proposed deep feature learning methodology effectively diagnoses and identifies bearing faults in metallic, hybrid, and ceramic bearings.
• The approach demonstrated adaptability and strong performance across two different electromechanical systems.
• Validation confirmed the methodology's suitability for condition-monitoring in systems with diverse bearing technologies.

Conclusions:

• The developed deep feature learning methodology offers a robust solution for bearing fault diagnosis across various bearing types.
• This data-driven approach significantly improves the adaptability and performance of condition-monitoring strategies.
• The study highlights the potential of advanced machine learning techniques in managing the complexities of modern rotary electromechanical systems.