• Proposes a multi-channel data fusion framework for intelligent fault diagnosis. • Addresses extreme data bias through a multilevel graph-guided learning strategy. • MultiGraph structures capture complementary features using diverse topological relationships. • A divergence-based clustering objective enhances feature separability and intra-category compactness. Fault diagnosis based on multi-channel data plays a crucial role in rotating machinery monitoring. By leveraging signals acquired from multiple sensors, more comprehensive fault-related information can be extracted, thereby improving diagnostic accuracy. This paper proposes a novel Multi-channel data fusion-enabled Multilevel Graph-guided Framework for Diagnosis (MSGFD) to address fault diagnosis under extreme data imbalance. First, an efficient preprocessing strategy is developed to transform multi-channel signals into structured representations suitable for graph-based learning. Subsequently, a MultiGraph construction mechanism is introduced to capture discriminative and complementary fault information through four distinct graph topologies. To address the challenge of limited supervision in extremely imbalanced scenarios, a multilevel learning architecture integrating a Graph Multilayer Perceptron (MLP) and a Graph Transformer is designed to jointly model local and global feature dependencies. Furthermore, a deep divergence-based clustering (DDC) loss is incorporated to enhance inter-class separability and intra-class compactness. Extensive experiments conducted under various imbalance settings demonstrate the robustness and effectiveness of the proposed method across multiple fault categories. The source code is publicly available at: https://github.com/Polimi-YuYue .
Yu et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: