276 / 2021-04-15 22:38:35

Normalized Conditional Variational Autoencoder for Imbalanced fault diagnosis of rolling bearings

rolling bearings,fault diagnosis,class imbalanced dada,conditional variational Auto-encoder,Focal loss

Machine condition monitoring and fault diagnosis

The distribution of health data monitored in the industrial field is mostly unbalanced. The amount of monitoring data for the normal condition is far more than monitoring data for fault conditions. Simultaneously, traditional intelligent diagnosis methods are mainly based on the assumption of balanced distribution of data categories. To this end, this paper designs a mechanical system category imbalance fault diagnosis framework based on Normalized conditional variational Autoencoder with Adaptive Focal Loss (NCVAE-AFL). The core of this diagnostic framework is to use the designed NCVAE to enhance the data’s feature learning ability, extract the multi-layer sensitive feature vector of the data, and improve the generalization performance of the diagnostic model. Simultaneously, a new adaptive focus loss (AFL) loss function is designed for NCVAE, which focuses training on a few samples of health conditions that are difficult to classify to balance the diagnosis difficulty of samples of different categories. Finally, the double-span rotor-bearing system fault simulation experiment platform verifies the effectiveness and superiority of the proposed NCVAE-AFL algorithm and its diagnostic framework. The diagnosis results show that the fault diagnosis framework's accuracy and stability based on the NCVAE-AFL algorithm are better than other latest methods when dealing with unbalanced fault data of rotating machinery.