An Emerging Fault Diagnosis Framework with Incremental Meta-Learning for Rotating Machinery

Recently., deep learning (DL) based intelligent fault diagnosis methods have attracted extensive interest due to its powerful automatic feature extraction ability and excellent generalization performance. However, existing data-driven fault diagnosis methods usually require a large amount of training data to help the model adapt to new diagnostic tasks, which is time-consuming and do not meet the requirements of online real-time fault diagnosis of streaming data in real industrial applications. Therefore, an Emerging Fault Diagnosis Framework with Incremental Meta-learning is proposed in this study for rotating machinery of intelligent fault diagnosis. In particular, a novel meta-update strategy with a dynamic weight factor is designed to alleviate the catastrophic forgetting of learned knowledge and to adapt to the emerging fault detection task successfully. Furthermore, Label Smoothing Regularization (LSR) is embedded into the developed framework to eliminate model overfitting. Extensive experiments are conducted on a classic bearing dataset and provide a convincing validation for the effectiveness of the proposed framework in incremental fault diagnostic tasks.