Residual Shrinkage Prototypical Network for Fault Detection of Propulsion Shaft System with Few Labeled data

In the field of engineering, the lack of labeled information for training data has been a troubling problem. To address this problem, this paper proposes a residual shrinkage prototypical network for fault detection of Propulsion Shaft System with few labeled data. The method constructs RSM to extract the hidden features of source and target domain data. The prototype calculation module is designed to calculate feature prototypes from the source and target domains. A similarity metric module is constructed for prototype feature learning to reduce domain alignment error. A threshold-based classifier is set to adaptive update parameters, which realizes fault diagnosis in few labeled data. To verify the performance of the proposed method, monitoring data of propulsion shaft system are collected under different operating conditions from actual operations to conduct cross-domain fault detection. The results show that the proposed method can achieve accurate detection under different operating conditions while reducing the dependence of the data-driven method on labeled data.