A review of few-shot fault diagnosis technologies for rotating machinery

Deep learning has achieved great successes in the field of rotating machinery fault diagnosis. Its excellent performance heavily relies on adequate training samples. However, it is very difficult to obtain sufficient training samples in industrial applications, which leads to poor generalization and low accuracy Therefore, few-shot fault diagnosis has gradually become an active research topic owing to their ability to mine fault-related information under the limited training data. In this paper, the latest achievements are reviewed and summarized for few-shot fault diagnosis of rotating machinery. Firstly, the definition and learning methods for few-shot fault diagnosis are described. Secondly, according to different technical principles, the existing few-shot fault diagnosis methods are categorized into five types: meta-learning, transfer learning, domain generalization, data augmentation, and self-supervised learning. Subsequently, the applications of these five approaches in rotating machinery fault diagnosis are summarized. Finally, the key ideas, advantages and limitations of these five methods are concluded, and the challenges of few-shot fault diagnosis methods are discussed.