Fault diagnosis is critical for ensuring the reliability and safety of industrial systems. Recently, graph convolutional networks (GCN) have gained significant attention for fault classification due to their ability to model complex dependencies in sensor signals. However, existing GCN-based approaches face two major limitations: (1) their adjacency matrices are typically predefined and fixed, failing to adaptively capture the varying importance of motor and gearbox nodes under different fault conditions, and (2) they lack comprehensive multi-scale sampling strategies, limiting their ability to construct informative time-series representations. To address these issues, we propose a multi-scale adaptive graph convolutional network (MSAGCN) for rapid fault classification. MSAGCN introduces an adaptive graph weighting mechanism, enabling the model to dynamically adjust node importance based on fault characteristics. Additionally, a multi-scale sampling aggregation strategy is incorporated to extract rich temporal features, enhancing the model’s discriminative power. Experimental results demonstrate that MSAGCN outperforms conventional GCN-based methods, achieving improved classification accuracy across various fault scenarios. The proposed framework establishes a robust baseline for fast and accurate fault diagnosis, highlighting the importance of adaptive graph structures and multi-scale feature extraction in industrial monitoring applications.
Zhao et al. (Mon,) studied this question.