81 / 2025-04-05 17:27:19

Fault Diagnosis for Pneumatic Control Valves Using a Novel Few-shot Learning Method

Pneumatic control valves, few-shot learning, fault diagnosis, graph neural networks, MAML

Pneumatic control valves are crucial for regulating fluid flow in industrial systems. Due to harsh operating conditions, they are prone to faults such as clogging and leakage, which can severely affect safety and efficiency. However, the scarcity of fault samples in industrial environments limits the applicability of existing deep learning methods, which rely heavily on large amounts of labeled data. To address this issue, a novel few-shot learning method based on GAC-GIN-MAML is proposed in this paper for fault diagnosis of pneumatic control valves. The proposed approach uses Fully Connected Graph (FCG) to construct the graph structure and leverages the GAC-GIN model to extract fault-related features. In addition, a task selection strategy is proposed to improve MAML, which is combined with GAC-GIN for fault diagnosis. Experimental results on the pneumatic control valve dataset from DAMADICS demonstrate the effectiveness and robustness of the proposed method. Compared to existing methods, significant improvements in fault diagnosis accuracy and generalization capability are achieved.