A new dynamic model and transfer learning based intelligent fault diagnosis framework for rolling element bearings...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent fault diagnosis for rolling element bearings is crucial.
• Current methods often require extensive historical data, posing a challenge in real-world applications.
• The small sample problem leads to overfitting and poor generalization in diagnosis models.

Purpose of the Study:

• To propose a novel intelligent fault diagnosis framework for rolling element bearings race faults.
• To address the small sample problem in bearing fault diagnosis.
• To enhance the generalization and accuracy of fault identification.

Main Methods:

• Utilizing a dynamic model to generate abundant and diverse simulation data for bearings.
• Applying transfer learning strategies with Convolutional Neural Networks (CNNs).
• Leveraging diagnosis knowledge from simulation data to real-world scenarios via parameter transfer.

Main Results:

• The proposed framework effectively generates large-scale simulation data.
• Transfer learning and parameter strategies in CNNs facilitate learning of more transferable features.
• A significant reduction in feature distribution discrepancy was observed.
• Fault identification performance was substantially enhanced.

Conclusions:

• The developed framework successfully overcomes the limitations of small datasets in bearing fault diagnosis.
• The integration of dynamic models and transfer learning significantly improves diagnostic accuracy and generalization.
• This approach offers a promising solution for intelligent fault diagnosis in rolling element bearings.