FCRNet: Fast Fourier convolutional residual network for ventilator bearing fault diagnosis

• Published in: PloS one Vol. 20; no. 7; p. e0327342
• Main Authors: Cao, Yu, Du, Yongzhi, Le, Likun, Li, Xiaoxue, Gao, Yanfang
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 11.07.2025; Public Library of Science (PLoS)
• Subjects: Accuracy; Algorithms; Artificial neural networks; Bearings; Bearings (Machinery); Condition monitoring; Deep learning; Equipment and supplies; Fault diagnosis; Fault location (Engineering); Fourier Analysis; Fourier transforms; Frequency dependence; Humans; Information processing; Inspection; Methods; Mine ventilation; Mining; Mining accidents & safety; Neural networks; Neural Networks, Computer; Noise levels; Propagation; Signal processing; Ventilation; Ventilators; Ventilators, Mechanical; Wavelet transforms; Working conditions; China
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0327342

This study presents FCRNet, a Fast Fourier Convolution Residual Network, tailored for fault diagnosis of mine ventilation bearings under complex operating conditions. By integrating residual learning with Fast Fourier Convolution (FFC), FCRNet employs a dual-branch architecture to effectively capture local spatial features and global frequency patterns. A Spectral Transformation (ST) module achieves unified processing of multi-scale spatial and frequency information by integrating local Fourier features (LFF), global fourier features (GFF), and local time-domain features (LF), overcoming the limitations of conventional convolutional approaches. The testing results on publicly available datasets and our self-built platform validate that the proposed method outperforms several existing fault diagnosis methods at various noise levels, providing strong support for the condition monitoring of mine ventilation.