A Novel Bearing Fault Diagnosis Method Based on Few-Shot Transfer Learning across Different Datasets.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent fault diagnosis relies heavily on large datasets of artificial simulation faults (ASF).
• Acquiring extensive real-world fault data is often impractical and costly.
• Simulated faults offer valuable diagnostic knowledge but require effective transfer.

Purpose of the Study:

• To develop a bearing fault diagnosis method using few-shot transfer learning across different datasets (cross-machine).
• To leverage knowledge from ASF for diagnosing bearings with natural faults (NF).
• To improve diagnostic performance with limited target domain samples.

Main Methods:

• Utilized a Siamese network framework for feature extraction.
• Employed a fault support set for comparison and adaptation.
• Fine-tuned the model using a minimal number of natural fault samples.

Main Results:

• The proposed method successfully learned diagnostic knowledge from diverse ASF datasets.
• Accurate identification of natural fault states in bearings was achieved.
• Demonstrated strong generalization and robustness across different datasets (CWRU, PU).

Conclusions:

• Few-shot transfer learning effectively bridges the gap between simulated and natural bearing faults.
• The method offers a practical solution for fault diagnosis with limited data.
• The approach avoids the need for secondary training, enhancing applicability.