An improved semi-supervised prototype network for few-shot fault diagnosis

Abstract

The collection of labeled data for transient mechanical faults is limited in practical engineering scenarios. However, the completeness of sample determines quality for feature information, which is extracted by deep learning network. Therefore, to obtain more effective information with limited data, this paper proposes an improved semi-supervised prototype network (ISSPN) that can be used for fault diagnosis. Firstly, a meta-learning strategy is used to divide the sample data. Then, a standard Euclidean distance metric is used to improve the SSPN, which maps the samples to the feature space and generates prototypes. Furthermore, the original prototypes are refined with the help of unlabeled data to produce better prototypes. Finally, the classifier clusters the various faults. The effectiveness of the proposed method is verified through experiments. The experimental results show that the proposed method can do a better job of classifying different faults.