A Novel Deep Transfer Learning Method for Intelligent Fault Diagnosis Based on Variational Mode Decomposition and...

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Deep learning excels in intelligent fault diagnosis but struggles with varying operating conditions between training and testing datasets.
• Signal preprocessing significantly impacts diagnostic model performance, posing a challenge for integration with transfer learning models.

Purpose of the Study:

• To propose a novel deep transfer learning method for intelligent fault diagnosis that addresses limitations of current methods.
• To enhance the accuracy, generality, and robustness of fault diagnosis models under diverse operating conditions.

Main Methods:

• Utilized Variational Mode Decomposition (VMD) to adaptively decompose signals, optimizing center frequencies and bandwidths for effective signal separation.
• Employed Efficient Channel Attention (ECA) to learn and fuse mode features post-VMD decomposition, enhancing feature representation.

Main Results:

• The proposed VMD-ECA based signal preprocessing and feature fusion module significantly improved the accuracy and generality of the transfer diagnostic model.
• The method demonstrated superior robustness and generalization performance compared to state-of-the-art techniques across various noise levels.

Conclusions:

• The novel deep transfer learning approach effectively integrates signal preprocessing with diagnostic models, overcoming challenges of domain shift.
• The VMD-ECA module offers a robust solution for intelligent fault diagnosis, enhancing model performance in real-world applications.