A Pseudo-Labeling Multi-Screening-Based Semi-Supervised Learning Method for Few-Shot Fault Diagnosis

• DOI: 10.3390/s24216907
• Publication Date: 2024-10-28T12:39:07Z
• Authors: Shiya Liu; Zheshuai Zhu; Zibin Chen; Jun He; Xingda Chen; Zhiwen Chen
• Keywords: Feature (linguistics); AdaBoost; Sample (material)

In few-shot fault diagnosis tasks in which the effective label samples are scarce, existing semi-supervised learning (SSL)-based methods have obtained impressive results. However, industry, some low-quality hidden collected dataset, can cause a serious shift model training and lead to performance of SSL-based method degradation. To address this issue, latest prototypical network-based SSL techniques studied. most scenarios consider that each sample has same contribution class prototype, ignores impact individual differences. This article proposes new based on pseudo-labeling multi-screening for bearing diagnosis. proposed work, strategy is explored accurately screen improving generalization ability network. addition, AdaBoost adaptation-based weighted technique employed obtain accurate prototypes by clustering multiple samples, deteriorated samples. Specifically, squeeze excitation block used enhance useful feature information suppress non-useful extracting accuracy features. Finally, three well-known datasets selected verify effectiveness method. The experiments illustrated our receive better than state-of-the-art methods.