Skip to content
Open access

Physics driven incremental meta-learning rolling bearing fault diagnosis under few-shot emerging conditions

Oct 2026 · Measurement science and technology · Vol 37 · 0 citations · 51 references
Physics

Abstract

Traditional deep learning fault diagnosis methods rely on sufficiently labeled training data, which limits their ability to adapt to the continual emergence of new fault data. Although continual learning methods can alleviate catastrophic forgetting to some extent, the model’s diagnostic performance is still significantly impacted under few-shot emerging fault data. To address these issues, this paper proposes a physics-driven incremental meta-learning method for rolling bearing fault diagnosis. First, the method uses a physics model to generate multi-condition bearing fault data, providing original training data for the meta-training stage. This approach effectively alleviates the problem of limited training data. Then, under the model-agnostic meta-learning framework, the model is meta-trained using the simulated data, enabling the model to rapidly adapt to few-shot emerging fault data under self-labeling conditions. Finally, the self-labeling fault data are combined with historical data stored in the memory buffer through prototype replay. Such a mechanism enables continual incremental diagnosis of bearing faults. The proposed method in this paper was validated using existing datasets. The experimental results show that under different situations of few-shot emerging fault data, the accuracy of the proposed method is higher than that of other methods.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.