Few-shot Fault Diagnosis Based on Supervised Contrast Learning

Abstract

In recent years, deep learning has performed extremely well in the field of fault diagnosis of rolling bearings due to its powerful feature extraction capability. However, in the case of a very small dataset of labelable samples, fault diagnosis still faces the status quo of insufficient feature learning and inaccurate fault category differentiation. This paper proposes a supervised contrast learning framework based on few samples. In the proposed framework, firstly, a small number of samples are passed through an encoder for feature extraction. Second, the similarity and difference of labeled samples are used to construct positive and negative sample pairs. Finally, the features learned by the encoder after contrast learning are passed through a classifier to achieve fault classification. The framework in this paper greatly reduces the time cost of manual preparation of labels. Experimental results show that the bearing fault diagnosis model proposed in this paper has high diagnostic accuracy.