Partial Transfer Learning Method Based on Inter-Class Feature Transfer for Rolling Bearing Fault Diagnosis.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Transfer learning in rolling bearing fault diagnosis often assumes consistent sample classes between source and target domains.
• Early-stage mechanical applications may lack complete fault class data, leading to label imbalance in the target domain.
• Outlier classes in the source domain, without target domain counterparts, can negatively impact feature alignment and transfer.

Purpose of the Study:

• To develop an innovative inter-class feature transfer approach for rolling bearing fault diagnosis.
• To address the challenge of label imbalance and outlier classes in transfer learning scenarios.
• To improve the accuracy and robustness of intelligent bearing fault diagnosis systems.

Main Methods:

• The proposed method utilizes label information to compute distribution discrepancies among shared classes.
• It specifically circumvents the interference of outlier classes from the source domain.
• The approach focuses on effective feature alignment for partial transfer tasks.

Main Results:

• Empirical evaluations on two rolling bearing datasets demonstrated superior performance compared to existing methods.
• The approach successfully mitigated the negative influence of outlier classes on feature transfer.
• Significant improvements in fault diagnosis accuracy were observed under partial transfer conditions.

Conclusions:

• The introduced inter-class feature transfer method offers a novel and effective solution for intelligent bearing fault diagnosis.
• It robustly handles label imbalance and outlier classes, enhancing transfer learning efficacy.
• This work provides a valuable contribution to the field of condition monitoring and predictive maintenance for rolling bearings.