基于融合知识迁移网络的变工况轴承故障模式识别

Fault Pattern Recognition of Bearing Under off Design Condition Based on Knowledge Transfer Network

针对基于迁移学习的故障诊断方法无法充分利用目标域数据,并且要求运行条件平稳,提出了一种基于融合知识迁移网络的变工况轴承故障模式识别方法。将输入瞬时转速作为工况信息输入到稀疏自动编码器中,从而充分利用目标域信息,使操作信息不必只利用局部振动数据集,而可以将整个操作信息纳入模型进行训练,并且通过模型训练大大降低了学习过程中负迁移的风险。然后利用深度卷积神经网络从原始振动中提取特征,通过两种知识迁移模型的结合,建立了融合知识迁移模型。最后,在滚动轴承实验测试台上的实验结果验证了该方法能够在变工况条件下实现有效的故障识别。

In view of the fact that the fault diagnosis method based on transfer learning could not make full use of the data in the target domain and required smooth operation conditions,a fault pattern recognition method of variable working condition bearing based on fusion knowledge migration network was proposed.The input instantaneous speed was input into the sparse automatic encoder as the working condition information,so that the target domain information could be fully utilized,and the whole operation information could be incorporated into the model for training instead of only using the local vibration data set,and the risk of negative transfer in the learning process was greatly reduced through the model training.Then,the deep convolution neural network was used to extract features from the original vibration,and the fusion knowledge transfer model was established by combining the two knowledge transfer models.Finally,the experimental results on the rolling bearing test-bed show that the method can achieve effective fault identification under variable working conditions.