Gearbox Fault Diagnosis for Wind Turbines based on Data Augmentation using Improved Generative Adversarial Networks

Abstract

Under realistic working conditions, the fault data of the gearbox of the wind turbines are difficult to balance with the healthy data. The imbalance of the data will affect the training results of the fault diagnosis model. To deal with such an imbalance, a fault diagnosis method with data augmentation is proposed for gearbox of the wind turbines. The method takes the spectrograms of the vibration signals as the input. Then we develop a framework of generative adversarial networks as the data augmentation model. Among the framework, the weight matrices of networks are all improved through spectrally normalization. Then the data augmentation model can be training more stably to generate samples with a high quality and diversity. Finally, taking the enhanced spectrograms as the training set, the fault diagnosis model is obtained based on the convolutional neural networks. The proposed method is verified to diagnose four working states. The results denote that the proposed method is effective in imbalanced data set.