Bearing Fault Diagnosis with Denoising Autoencoders in Few Labeled Sample Case
Yuan Zeng, Xiaojun Wu, Jinlong Chen · 2020
The application of transfer learning in the fields of computer vision and natural language processing has been successful. It turns out that pre-training some parameters in the network with unsupervised tasks related to downstream tasks is very useful for subsequent task model training. This approach can reduce the dependence of model training on labeled data to a certain extent, which is beneficial to the application of deep learning algorithms. This paper mainly proposes a bearing fault diagnosis model with a small number of labeled samples. The model is trained through pre-training and fine-tuning. It can use a large amount of unlabeled data to obtain sufficient data representation. Finally, fine-tuning is achieved on small sample fault sets Accuracy. At the same time, several optimizations are proposed on the dassic AE fault diagnosis method, and the effectiveness of the optimization content is verifiedin the experimental stage.