To address the challenges of insufficient scale normalization, limited time–frequency localization, and ineffective multi-scale feature extraction in the intelligent fault diagnosis of rotating components under varying operating conditions, we propose a novel convolutional neural network, termed DG-FuseNet. The proposed method was validated on real-world datasets from train vehicle vibration signals and aero-engine systems, achieving diagnostic accuracies of 99.76% and 94.32%, respectively. Compared with eleven advanced intelligent models, DG-FuseNet demonstrated faster convergence, higher diagnostic accuracy, strong robustness against interference, and superior generalization capability. These results indicate that DG-FuseNet outperforms existing approaches in complex industrial scenarios, highlighting its excellent performance and stability.
Zhang et al. (2026) studied this question.