Deep Transfer Learning for Imbalanced Fault Diagnostics
Intelligent fault diagnostics suffers from class-imbalance and domain variance problems simultaneously. Traditional class-imbalance learning or transfer learning methods can not guarantee accuracy and generalization for fault diagnosis on target datasets. Thus, a deep transfer learning framework for imbalanced fault diagnostics should be developed. Three tasks should be studied:
- Hybrid imbalance learning based on adaptive oversampling and cost-sensitive learning methods.
- ImbaTran-Net: a deep transfer network for imbalanced datasets
- ImbaTran-Net for multiple source domain transfer tasks: alleviate negative transfer effect on imbalance transfer tasks