An Integrated Multitasking Intelligent Bearing Fault Diagnosis Scheme Based on Representation Learning Under...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate bearing fault diagnosis is critical for mechanical system reliability.
• Imbalanced datasets (more healthy than faulty data) pose challenges in real-world scenarios.
• Existing methods struggle with the integrated tasks of fault detection, classification, and identification.

Purpose of the Study:

• To propose an integrated, multitasking intelligent bearing fault diagnosis scheme.
• To address imbalanced sample conditions using representation learning.
• To achieve bearing fault detection, classification, and unknown fault identification.

Main Methods:

• Developed a modified denoising autoencoder with a self-attention mechanism for bottleneck layer (MDAE-SAMB) for unsupervised fault detection using only healthy data.
• Employed transfer learning based on representation learning for few-shot fault classification.
• Integrated these methods for comprehensive bearing fault diagnosis.

Main Results:

• The MDAE-SAMB effectively detects bearing faults using healthy data alone.
• Few-shot learning achieved high-accuracy online fault classification.
• Unknown bearing faults were successfully identified using known fault data.
• The scheme demonstrated applicability on both RDER and public bearing datasets.

Conclusions:

• The proposed integrated multitasking scheme offers a robust solution for bearing fault diagnosis under imbalanced conditions.
• Representation learning and self-attention mechanisms enhance detection and classification accuracy.
• The approach effectively handles detection, classification, and identification of known and unknown faults.