Normalized Conditional Variational Auto-Encoder with adaptive Focal loss for imbalanced fault diagnosis of Bearing-Rotor system

Abstract

The distribution of the health data monitored from mechanical system in the industries is class imbalanced mainly. The amount of monitoring data for the normal condition is far more than the monitoring data for different fault conditions. Simultaneously, traditional intelligent diagnosis methods are primarily based on the assumption of the balanced distribution of data categories. To this end, this paper designs a new class imbalanced fault diagnosis framework of the bearing-rotor system based on Normalized Conditional Variational Auto-Encoder with Adaptive Focal Loss (NCVAE-AFL). The critical point of this diagnostic framework is to use the designed NCVAE algorithm to enhance the data’s feature learning ability. The multi-layer sensitive feature vectors of the data can be extracted, the generalization performance of the diagnostic framework is further improved. Meanwhile, a new Adaptive Focus Loss (AFL) function is designed for NCVAE model, which focuses training on a few samples of health conditions that are difficult to classify and balance the diagnosis difficulty of the numerous sample categories. Finally, the double-span bearing-rotor system fault simulation experiment platform verifies the effectiveness and superiority of the proposed NCVAE-AFL algorithm and its diagnostic framework. The diagnosis results demonstrate that the proposed diagnosis framework’s accuracy and stability are better than other latest methods when dealing with the class imbalanced fault data of mechanical systems.