Fault Diagnosis of Mechanical Transmission Components with Siam-GNN Under Few-Shot Noisy Conditions

Fault diagnosis of mechanical transmission components is challenged by limited fault samples and noise interference. Deep learning methods often overlook the implicit topological relationships between samples. Graph neural networks (GNNs) efficiently aggregate feature information of neighboring nodes based on the dependencies and topological relationships among samples, demonstrating strong reasoning capabilities. However, determining node adjacency via K-nearest neighbor (KNN) in traditional GNNs can lead to node misconnections under few-shot conditions with noise interference, resulting in erroneous feature aggregation. Therefore, this paper proposes an inductive fault diagnosis method based on a Siamese Graph neural network (Siam-GNN). Firstly, this paper introduces the Siamese network to determine node adjacency relationships innovatively, optimizing the quality of association graph by increasing distances between nodes of the same category and decreasing distances between nodes of different categories. Subsequently, the classic contrastive loss function is improved to enhance the robustness under noisy conditions. Finally, ablation and comparative experiments are conducted on rolling bearings and planetary gearboxes based on the proposed fault diagnosis method, demonstrating the proposed method's stronger robustness, higher diagnostic accuracy and good generalization performance across both types of transmission components.