Few-Shot Fault Diagnosis Method For Rotating Machinery Based on Meta-Transfer Learning

Fault diagnosis methods based on deep learning have become a research hotspot because they can be classified adaptively, but these methods usually need sufficient training data to achieve. In industrial applications, it is often difficult to collect sufficient annotation data to train models. In addition, rotating machinery and equipment may have unexpected new faults during operation. When a new fault occurs and there are few samples available, it is a challenge to train a reliable model with all types of data. Therefore, starting from meta-transfer learning, this paper introduces Model-Agnostic Meta-Learning (MAML) and Fine tuning to solve the small sample problem when a new fault occurs. The data sets under two different new fault situations are constructed respectively, and experiments are designed, and the advantages of the method in this paper are verified by comparing with the five methods.