As a core component of rotating machinery, bearing health status is directly related to the operational stability and safety of the equipment. In the field of bearing fault diagnosis, although vibration signals are commonly used as monitoring data sources, their contact-based acquisition method is susceptible to constraints imposed by installation conditions and hardware parameters. To address this issue, this paper proposes a non-contact diagnostic method based on acoustic signals and constructs the 1dCMPR-BiSEnet model. In this study, six bearing operating conditions are defined, covering typical failure modes such as inner-race, outer-race, and cage faults. On this basis, a bearing acoustic dataset is constructed. Relying on the collected acoustic data, fault diagnosis experiments are conducted with the proposed method. The experimental results reveal that the diagnostic accuracy of this method reaches 99.97% under different rotational speeds, with an average AUC of 1.0 and an average AP of 0.9996. It delivers better overall performance than comparative algorithms and presents satisfactory robustness under high-noise environments. Furthermore, verified by the University of Ottawa Bearing Dataset, the method achieves a generalization accuracy of 98.06% in vibration signal scenarios. The high accuracy of the proposed method demonstrates that this study can provide valuable insights for relevant research in the field.
Liu et al. (Tue,) studied this question.