Research of turbine rotor fault diagnosis based on improved auxiliary classification generative adversarial network

The rotor is an important part of the turbine, but the vibration information of the rotor is not easy to be extracted, which leads to the lack of its vibration data. In this paper, a data augmentation method for assisting in turbine rotor fault diagnosis, the Auxiliary Classifier Wasserstein Generative Adversarial Network with Self-Attention Mechanism (SA-ACWGAN), is improved to solve this problem. The Auxiliary Classification Generative Adversarial Network (ACGAN) as an architecture ensures the balance of the generated data, the incorporated Wasserstein distance ensures the accuracy of the feature extraction, and the Self-Attention Mechanism module enables the generator and the discriminator to consider both the local and global features in the feature extraction. Experiments are conducted on different rotor datasets. The results show that the method is effective in identifying faults in turbine rotors, with accuracy higher than 97% for both datasets.