Progressive orthogonal matching pursuit and adaptive modal screening for fault diagnosis of rotating machinery components using acoustic signals

Abstract

To overcome limited placement options of contact-based vibration sensors in detecting rotating machinery, a novel fault diagnosis method is proposed, integrating non-contact measurement with Progressive Orthogonal Matching Pursuit and Adaptive Modal Screening (POMP-AMS). This approach aims to boost diagnostic accuracy by extracting key components from acoustic signals. It addresses weak fault signatures and noise issues by using a POMP-AMS-based noise reduction and feature enhancement technique. Utilizing cross-correlation coefficients and Fourier bases, it performs correlation orthogonal matching pursuit to eliminate extreme noise. Adaptive modal filtering further enhances fault features. Selected modes construct an initial matrix for noise reduction and sparse representation, resulting in filtered and reconstructed feature signals. Validated with the Ottawa dataset, the method surpasses Fast Kurtogram, OMP, and GSL in extracting hidden faults and assessing bearing health, offering low-cost data acquisition and high early fault recognition, ideal for practical engineering monitoring and diagnosis.