A Rotating Machinery Fault Diagnosis Method Based on Recurrence Characteristics and Model-Agnostic Meta-Learning Under Small Sample Conditions

To solve the problem that traditional deep learning methods for rotating machinery (RM) are difficult to learn effective fault features and consistent prior knowledge when the fault samples are insufficient and large differences in data distribution. We propose a fault diagnosis method for RM under small sample conditions based on the recurrence characteristics (RC) common to RM and combine it with model-agnostic meta-learning (MAML). Experiments on the CWRU bearing dataset and the NGW planetary gearbox dataset are employed to verify the accuracy and effectiveness. The results show that under the conditions of insufficient fault samples and large differences in data distribution, our method can effectively improve the accuracy of fault diagnosis.