Imbalanced Real-Time Fault Diagnosis Based on Minority-Prioritized Online Semi-Supervised Random Vector Functional Link Network
Pengyu Han, S.-S. Chen, Zeyi Liu, Xiao He
Abstract
Industrial real-time fault diagnosis is vital to ensure efficient and safe production. In the literature, existing methods usually do not systematically consider some realistic constraints in dealing with the above problem, such as real-time model update, sample imbalance, and high cost of labeling. In this paper, a minority-prioritized online semi-supervised random vector functional link network approach, termed MPOS-RVFL, is proposed to cope with the above issues. Specifically, the pseudo-labeling technique is introduced to fully exploit the information from unlabeled samples in the online data stream. In this context, the approach incorporating minority anchors prioritization, minority weight, and pseudo-label is developed to enhance the model’s capability in accurately identifying minority samples. Several experiments with a real-world gearbox fault dataset are conducted to verify the practicality of MPOS-RVFL. The results demonstrate that the proposed method outperforms the existing state-of-the-art approaches. The source code is available at https://github.com/THUFDD/MPOS-RVFL.