Acoustic Fault Diagnosis Method for Rotating Machinery Based on Collaborative Perception Information Aggregation Guidance Network

The non-contact and directional nature of acoustic fault diagnosis offers a distinct advantage for fault detection in complex environments, ensuring the safe and stable operation of rotating machinery. However, challenges such as complex working conditions and data imbalance significantly hinder the effectiveness of acoustic-based diagnostic systems. To address these issues, this paper proposes a Collaborative Perception Information Aggregation Guidance Network (CPAGN). In the proposed method, a Stem layer is utilized at the front end to extract nonlinear acoustic features. Subsequently, the collaborative perception layer integrates local frequency information extraction, multi-receptive field temporal information fusion, and redundancy reduction, thereby capturing the time-frequency variations of acoustic signals under complex operating conditions and enhancing condition awareness. Finally, inter-layer feature correlation computation is employed to generate global channel guidance, facilitating the effective interaction and fusion of global information. The CPAGN successfully extracts critical acoustic features and identifies essential fault information, demonstrating remarkable robustness and generalizability in handling data imbalance. To validate the effectiveness of CPAGN, comparative experiments were conducted on two datasets, comparing it with several existing deep learning methods. The results indicate that CPAGN excels at capturing fault information from acoustic signals, achieving superior diagnostic performance in complex working conditions and imbalanced data scenarios.