A semi-supervised prototypical network with dual correction for few-shot cross-machine fault diagnosis

• DOI: 10.1088/1361-6501/adc7d0
• Publication Date: 2025-04-01T22:50:02Z
• Authors: Guigang Liu; Kairong Gu; Haifeng Jiang; Jianhua Zhong; Jianfeng Zhong
• Keywords: One shot

Abstract Meta-learning has been widely applied and achieved certain results in few-shot cross-domain fault diagnosis due to its powerful generalization robustness. However, existing meta-learning methods mainly focus on within the same machine, ignoring fact that there are more significant domain distribution differences sample imbalance problems between different machines, leading poor diagnostic performance. To address these issues, this paper proposes a semi-supervised prototypical network with dual correction (SPNDC). First, dual-channel residual is utilized comprehensively extract features, capturing both deep shallow information. Then, correct by weighting features adding shift term support set samples query samples, respectively, diminish intra-class bias extra-class bias. Meanwhile, regularization introduced into model balance among class prototypes, enhancing distinctiveness. Finally, cross-machine experiments conducted three datasets validate effectiveness of method. Additionally, an interpretability analysis using gradient-weighted activation mapping (Grad-CAM) technique discern primary regions classification tasks.