Few-shot rolling bearing fault classification method based on improved relation network
| Autor: | Shouqiang Kang, Xintao Liang, Yujing Wang, Qingyan Wang, Chunyang Qiao, V I Mikulovich |
|---|---|
| Rok vydání: | 2022 |
| Předmět: | |
| Zdroj: | Measurement Science and Technology. 33:125020 |
| ISSN: | 1361-6501 0957-0233 |
| DOI: | 10.1088/1361-6501/ac8ca6 |
| Popis: | In practical applications, it is difficult to obtain enough fault samples to train a fault classification model for rolling bearings, and the specifications of bearings used in different mechanical equipment may be different. The diagnosis model trained on a certain specification of bearing may not be applicable to another specification. To solve the above problems, a few-shot rolling bearing fault classification method is proposed, based on an improved relation network (RN). First, a Fourier transform is applied to the vibration signals of different specifications of bearings. The data from different specifications are divided into a meta-train set and a meta-test set according to the meta-learning training strategy, and each set is further divided into a support set and a query set. Second, an improved RN is built. The residual shrinkage module and the scaled exponential linear unit activation function are introduced into the embedding module of the RN. The improved embedding module is used to extract the sample features of the support set and query set, and then the features of the two are combined and input into the relation module to get the relation score. The query set samples are classified according to the score. Finally, the rolling bearing fault classification model is obtained after multiple episodes. The experimental results show that, compared with the partial transfer learning and meta-learning methods, the proposed method only needs a few or even a single sample to achieve the fault classification of different specifications of rolling bearings under different loads. In the case of one-shot, the average classification accuracy can reach 93.3%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|