Lightweight Knowledge Distillation-Based Transfer Learning Framework for Rolling Bearing Fault Diagnosis

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Fault diagnosis across devices presents challenges due to significant data distribution differences.
• Existing transfer learning methods inadequately address cross-device fault diagnosis.
• Balancing computational resources and diagnostic accuracy is crucial for practical applications.

Purpose of the Study:

• To propose a knowledge distillation-based lightweight transfer learning framework for rolling bearing fault diagnosis.
• To address the limitations of current transfer learning approaches in cross-device fault diagnosis.
• To enhance diagnostic accuracy while minimizing computational and parameter overhead.

Main Methods:

• A deep teacher-student model using variable-scale residual networks was developed to learn domain-invariant features.
• A knowledge distillation framework with a temperature factor was employed to transfer knowledge from a large teacher model to a smaller student model.
• Multi-kernel domain adaptation was utilized to minimize feature distribution distance between source and target domains in Reproducing Kernel Hilbert Space (RKHS).

Main Results:

• The proposed framework effectively learned domain-invariant features for fault classification across different devices.
• Knowledge distillation significantly reduced the computational and parameter overhead of the diagnostic model.
• The method demonstrated effectiveness and applicability in scenarios with incomplete data across device types.

Conclusions:

• The developed framework offers a viable solution for cross-device fault diagnosis of rolling bearings, balancing accuracy and efficiency.
• The integration of knowledge distillation and domain adaptation provides a robust approach for handling data discrepancies between devices.
• Validation through engineering cases confirms the practical applicability of the proposed method in real-world operational settings.