The interpretable health condition monitoring method of gear transmission system embedded in feature cloud-guided hypergraph structure under the extremely small-sample background

Abstract

Addressing the condition monitoring challenges faced by gearboxes under extremely small-sample conditions, this study proposes an interpretable hypergraph discriminative embedding approach based on feature cloud augmentation. Specifically, a supervised feature learning framework based on hypergraphs is first established in this study, which effectively maps the raw health features of gearboxes into an intuitive and low-dimensional manifold space. This transformation facilitates the extraction of highly discriminative feature representations for various health conditions. Furthermore, to tackle the instability or even failure of subspace solutions during the construction of the low-dimensional manifold space in hypergraphs due to the extremely small-sample challenge, a feature cloud-based augmentation method is innovatively designed in this study. By constructing an augmented feature matrix, this strategy fundamentally alleviates the small sample constraint, thereby ensuring the effectiveness and stability of hypergraph learning. To comprehensively validate the effectiveness of the proposed method, this study conducted extensive tests and verifications on both the fault dataset from a drivetrain diagnostics simulator and practical fault monitoring data from wind turbine gearboxes. The experimental results demonstrate that the method can accurately monitor the health conditions of gearboxes under extremely small-sample conditions, significantly enhancing the reliability and safety of equipment operation.