关键词: 轴承故障诊断, 负载不平衡, 小样本, 短时傅里叶变换, 梯度惩罚Wasserstein距离生成对抗网络, 自注意力卷积神经网络

Abstract: Aiming at the problems that the bearing vibration signals were easily disturbed by unbalanced load and the small number of bearing fault samples, a bearing fault diagnosis method based on WGAN-GP and SeCNN was proposed. The bearing vibration signals were processed by short-time Fourier transform to get the time-spectrum samples that were easy to be processed by WGAN-GP, which were divided into training set, validation set and test set. Then the training set was inputted into WGAN-GP for adversarial training, new samples were generated with similar distribution to the training samples, and added to the training set to expand the training set. The expanded training set was input into SeCNN for learning, and the trained model was applied to the test set and output the fault recognition results. The analysis of the CUT-2 platform unbalanced load bearing data was carried out, and the experimental results show that the proposed method may accurately and effectively classify the bearing faults.

Key words: bearing fault diagnosis, unbalanced load, small sample, short-time Fourier transform, gradient penalty Wasserstein generative adversarial network（WGAN-GP）, convolutional neural network with self-attention mechanism（SeCNN）