Physics-Inspired Sparse Voiceprint Sensing for Bearing Fault Diagnosis

Voiceprint sensing (VS) technique provides a novel and low-intervention tool for bearing condition monitoring. However, it remains a challenging task to detect the unique acoustic patterns generated from incipient bearing faults, especially under low signal-to-noise ratio conditions. Motivated by this limitation, a physics-inspired sparse VS is innovatively proposed for bearing fault diagnosis. In this article, inspired by the physical structure of the acoustic signals emanating from bearings, a group spike-and-slab prior is first designed to sharp fault features. Afterward, a generalized sparse Bayesian learning framework is constructed to recover the fault-induced sparse impulses from a probabilistic perspective. Finally, the superiority of the proposed method is validated through simulation analyses and experimental studies. Compared with state-of-the-art methods, the proposed approach still achieves a significant performance improvement rate of 93.8% even under noisy scenarios.