An Improved Metalearning Framework to Optimize Bearing Fault Diagnosis under Data Imbalance

The intelligent diagnosis of rotating machinery with big data has been widely studied. However, due to the variability working conditions and difficulty in marking fault samples, it is difficult obtain enough high-quality for training bearing models practical industrial application scenarios. Aiming at problem imbalance caused by lack a novel metalearning method (MOFD) proposed get solution under imbalance. Firstly, order enhance variety Feature Space Density Adaptive Synthetic Minority Oversampling Technique (FSDA-SMOTE) this paper, which takes density difference minority samples spatial domain within class as constraint local neighbor similarity generate new augmentation. In addition, strengthen model’s learning ability performance limited residual-attention convolutional neural network (RA-CNN) was constructed identify deep features signals, strategy based on parameter gradient optimization applied RA-CNN refining process model. Finally, reliability verified through experimental analysis public dataset.