Answers are grounded in this paper and cite their sources.
High-speed train bearings operate under strong noise with extremely scarce fault labels, whereas laboratory testbeds provide abundant labeled data but with a severe domain shift. This paper proposes an interpretable fault diagnosis framework integrating mechanism-based feature extraction, a wide first-layer kernel CNN (WDCNN), and CORAL transfer learning. A 15-dimensional feature set (11 time-domain indicators and 4 fault-characteristic-frequency band energies) is constructed from bearing kinematics and impact dynamics, and a three-stage scheme (common-feature extraction, CORAL alignment, target-domain fine-tuning) bridges the testbed-to-train gap. The model achieves source-domain precisions above 0.94, outperforming SVM, RF, GBDT, and a standard DNN, and all predictions on 16 unlabeled target samples reach confidence no lower than 0.78. SHAP and UMAP analyses confirm that the decisions are consistent with bearing fault physics, providing an accurate, robust, and interpretable solution for train bearing monitoring.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: