Abstract
In Industry 4.0 intelligent manufacturing, rolling bearings serve as core components of rotating machinery. Their health status directly impacts the safety and reliability of entire manufacturing systems. However, existing fault diagnosis methods face critical challenges in noisy environments, including layer-wise feature information attenuation, insufficient multi-scale feature capture, and limited noise robustness. Such limitations create an urgent need for high-precision and robust deep learning diagnostic techniques. To address these challenges, this study proposes Channel-Time Parallel Attention Network (CT-ParaNet). The network innovatively designs a channel-time parallel attention mechanism that synchronously processes channel and temporal feature correlations to effectively solve information degradation in serial structures. The network constructs multi-scale parallel attention residual blocks using parallel multi-branch architecture with adaptive gating mechanisms to capture and fuse multi-scale fault features. Additionally, it establishes a serial-parallel hybrid processing architecture that systematically integrates parallel attention mechanisms with multi-scale feature extraction modules for hierarchical and parallel fine processing of fault signals. Experimental results on two independent bearing fault datasets show CT-ParaNet achieves accuracies of 98.53% and 98.29%, improving by 15.84 and 16.15% points over traditional methods respectively. Under extreme - 5dB signal-to-noise ratio (SNR) conditions, accuracies remain above 87% across Gaussian white noise, impulse noise, and colored noise environments. With only 0.1 training data ratio, accuracies exceed 92% on both datasets. CT-ParaNet significantly enhances accuracy and robustness of bearing fault diagnosis in noisy environments, providing important technical support for intelligent manufacturing equipment health monitoring.