15 / 2025-04-15 10:56:03

A parallel channel bearing fault diagnosis framework based on frequency information

Fault diagnosis, Attention mechanism, Frequency information, Dual channel, Information fusion.

Special Sessions > SSL 05 Condition monitoring, fault diagnosis and intelligent maintenance of mechanical equipment

Bearings as the core component of rotating machinery, effective fault diagnosis methods play a crucial role in ensuring the secure functioning of equipment. Frequency components are the key features reflecting the operating state of bearings, and a parallel channel network framework is proposed to address the problem that the health state of rolling bearings is difficult to be comprehensively characterized by a single time domain information. The frequency information is used to guide the two channels to extract features individually to make a diagnosis, and the diagnostic results are fused. The time domain channel works by alternating between convolutional and wavelet packet attention mechanism (WPAM) in order to enhance the network's ability to extract fault features from different frequency band components. Based on the bearing fault frequency characteristics, a network framework for multi-feature extraction is proposed for the frequency domain channel, with the heterogeneous perceptual convolution module (HPCM) and the global correlation focusing module (GCFM) to realize diversified capturing of frequency spectral information and long-range dependency modeling, respectively. The final experiment validates the efficacy of the proposed approach in enhancing diagnostic accuracy.