变工况及小样本情况下滚动轴承故障迁移学习方法综述

准确预测滚动轴承损伤类型及剩余寿命归于各类旋转机械运行的可靠性和安全性具有重要意义。迁移学习方法通过实验数据使网络模型学习到相关故障类型的知识，训练好的模型可以直接应用到实际工业生产问题当中。突破了传统深度学习方法所需大量有标签数据、模型使用有局限性、通用性差的局限。首先对不同类型的迁移学习方法在滚动轴承故障诊断中的应用进行分析，归纳总结面对不同情况下的迁移学习方法。其次，针对变工况、小样本及一些其他情况下的问题进行总结分析。最后，给出了滚动轴承的迁移学习发展趋势。

A review of rolling bearing migration learning under variable operating conditions and small samples

Accurate prediction of rolling bearing damage types and remaining life is important for the reliability and safety of operation of all types of rotating machinery. The transfer learning method enables the network model to learn the knowledge of relevant fault types through experimental data, and the trained model can be directly applied to actual industrial production problems. It breaks through the limitations of traditional deep learning methods that require large amount of labeled data, limited use of models, and poor generality. Firstly, we analyze the application of different types of transfer learning methods in rolling bearing fault diagnosis and summarize the transfer learning methods in the face of different situations. Secondly, the problems of variable working conditions, small samples and some other situations are summarized and analyzed. Finally, the development trend of transfer learning for rolling bearings is given.