变工况下基于联合适配与对抗学习的滚动轴承故障诊断

Rolling bearing fault diagnosis based on joint adaptation and antagonism learning under variable working conditions

针对在一些假设条件下采用深度卷积神经网络进行变工况场景下基于深度学习模型的轴承故障诊断,其诊断性能会大大降低。本文提出了多尺度卷积联合适配对抗网络(Multi-scale convolution joint adaptive countermeasure network,MSCJACN),即在MS-1DCNN模型中加入了迁移学习算法。通过特征提取模块提取通用特征,在特征提取器的最后两层全连接层上利用联合最大均值差异(Joint maximum mean difference,JMMD)对具体特征进行适配,同时在特征提取器后加域判别器,帮助网络提取域的不变特征。在12种迁移学习任务上进行消融实验,并与其他迁移学习算法进行对比实验,结果表明,MSCJACN在变工况故障诊断场景更具优势。

Under several assumption and variable working conditions,rolling bearing fault diagnosis based on deep learning model used deep convolution neural network,its diagnostic performance will be greatly reduced.So a multi-scale convolution joint adaptive countermeasure network(MSCJACN)is proposed in this paper,and it adds the migration learning algorithm to the MS-1DCNN model.The general features are extracted by feature extraction module,and the specific features are adapted by JMMD on the last two fully connected layers of feature extractor.At the same time,a domain discriminator is added after the feature extractor to help the network extract domain features invariant.The ablation experiments are carried out on 12 kinds of transfer learning tasks,and then and compared with other transfer learning algorithms.The results show that MSCJACN has more advantages in the fault diagnosis scenarios under variable working conditions.