基于迁移成分分析的传动轴承故障诊断研究

Research on transmission bearing fault diagnosis based on transfer component analysis

针对深度学习模型的传动故障诊断样本量不足的问题,提出了一种基于迁移成分分析和卷积神经网络的轴承故障诊断方法。对采样的轴承数据依据转速进行分帧形成数据集,将数据映射到高维希尔伯特空间,基于数据边缘分布差异最小原则构建迁移特征集,再将特征集导入多层卷积神经网络等多种判别器实现轴承的跨工况故障诊断。结果表明:搭建的模型上具有较高的跨工况域故障诊断准确率,可满足轴承故障诊断的要求,为跨工况条件下轴承的故障检测提供新的思路。

To solve the problem of insufficient sample size for transmission system fault diagnosis using deep learning model,a bearing fault diagnosis method based on transfer component analysis and convolutional neural network was proposed.The sampled bearing data is divided into frames according to the rotational speed to form the data set,the data is mapped to the high-dimension Hilbert space,and the migration feature set is constructed based on the principle of minimum difference in data edge distribution,and then the feature set is imported into the Convolutional Neural Network and other discriminators realize the fault diagnosis of bearing across working conditions.The results show that the model has a high cross-working condition fault diagnosis accuracy,which can meet the requirements of bearing fault diagnosis,and provide a new idea for bearing fault detection under cross-working condition.