Merge Multiscale Attention Mechanism MSGAN-ACNN-BiLSTM Bearing Fault Diagnosis Model

To solve the problem that the sample of rolling bearing in actual working condition is seriously imbalanced, which leads to the poor performance on accuracy and generalization of fault diagnosis model. In this paper, A multi-scale bearing fault diagnosis model MSGAN-ACNN-BiLSTM with progressive generation and multi-scale attention mechanism is proposed for imbalanced data. Firstly, the original imbalanced fault samples are transformed into multi-scale frequency domain samples and input into the multi-scale generative adversarial network for training. After the network reaches Nash equilibrium, the progressive generated multi-scale fault samples are mixed into the original imbalanced samples, so as to solve the problem of serious imbalance data in actual conditions. Then, the re-balanced multi-scale datasets is input into the diagnostic model for training, which can extract multi-scale global and local feature information and improve the performance of the model, so as to realize the accurate classification of bearing fault diagnosis under imbalanced data. This experiment is based on the data set of UConn and CWRU. The experimental results show that the performance of the generated data quality and diagnosis accuracy of the model in each dataset is higher than other comparison methods, which proves the stability and effectiveness of the model.