【摘要】 针对目前现有轴承故障诊断方法对不平衡数据集中的少数类诊断准确率低的问题，提出了不平衡数据集下基于重要性加权自编码器(Importance Weighted Auto-encoder, IWAE)的轴承故障诊断方法。首先通过少数类的样本数据来训练IWAE网络，将生成的样本数据加入到原始数据集中，得到平衡后的数据集；然后引入深度学习方法作为诊断网络，将平衡后的数据集直接输入诊断网络中，自适应的学习故障特征，实现故障分类。为了增强诊断网络的准确率，使用一维多尺度卷积神经网络进行故障诊断。大量的定性定量实验表明，所提出的方法在不平衡比为1/7时，少数类诊断的准确率已经能够达到98.90%,均优于其他现有模型，并且拥有较好的收敛性和泛化性。更多还原

【Abstract】 Aiming at the low accuracy with unbalanced data sets in existing bearing fault diagnosis methods, we proposed a bearing fault diagnosis method based on importance weighted auto-encoder(IWAE) in unbalanced data sets. It was trained by minority samples, and the generated samples were added into original data sets to obtain balanced data sets. Then, deep learning method was used as diagnose network, and the balanced data sets were fed into it as input, so as to adaptively learn fault characteristics and realize fault classification. A large number of qualitative experiments showed that when the imbalance rate was 1∶7, the method could correctly classify the balanced samples, and the accuracy rate was 98.90%. Based on various imbalance ratios, the proposed method had better convergence and generalization than other existing models.更多还原