Abstract: Deep learning has been widely used in the field of rotating machinery fault diagnosis nowadays.To improve the diagnosis effect of the deep learning oriented to a large number of unlabeled data and variable working conditions,a network model combining the feature learning ability of deep learning and the generalization ability of transfer learning was constructed.The deep shrinkage residual network was constructed by adding soft thresholds to extract the characteristics of bearing vibration data under noise redundancy.Then,the Joint Maximum Mean Deviation (JMMD) criterion and Conditional Domain Adversarial (CDA) learning domain adapting network were used to align the source and target domains.At the same time,adding Transferable Semantic Augmentation (TSA) regular items improved alignment performance between classes.The proposed model was verified by three kinds of experiments:variable load,variable speed and variable noise,and the result proved that the proposed method could overcome the shortcomings of traditional deep learning and shallow transfer learning algorithms.

Key words: intelligent fault diagnosis, variable working conditions, deep residual networks, unsupervised transfer learning

摘要： 目前,深度学习相关技术在旋转机械故障诊断领域得到了广泛应用。为提升深度学习算法在面对大量未标注数据和变工况运行方式下的诊断效果,构造了融合深度学习的特征学习能力与迁移学习的泛化能力的网络模型。通过添加软阈值构建深度收缩残差网络提取噪声冗余下的轴承振动数据的特征信息;采用联合最大平均偏差准则和条件对抗学习域适配网络对齐源域和目标域,同时添加正则项提高类间对齐性能;通过变负荷、变速度与变噪声三种实验设置验证了模型的有效性。实验结果证明,该方法能够有效克服传统深度学习和浅层迁移学习算法的不足。

关键词: 故障诊断, 变工况, 深度收缩残差网络, 无监督深度迁移学习