A Novel Singular and Harmonics-to-Noise Ratio Deconvolution for Fault Diagnosis in Rotating Machinery

Blind deconvolution enhances impulsive features without prior information and adapts to non-stationary signals, making it valuable for rotating machinery health monitoring. However, it faces two key challenges: 1) Under low SNR conditions, noise, and narrowband interference may be amplified, masking fault signals, and causing false detections. 2) Accurate estimation of the periodicity of fault impulses is crucial, but speed fluctuations, load variations, and modulation effects can lead to errors, resulting in misalignment of features and energy dispersion. To address these issues, a singular and harmonic noise ratio deconvolution (SHNRD) method is proposed in this work. First, an optimal singular spectrum decomposition (OSSD) is introduced to achieve non-uniform partitioning of fault-related frequency bands, effectively suppressing noise while preventing fault information leakage. Then, a harmonic noise ratio (HNR)-based adaptive filter length optimization strategy is designed to balance redundancy and information loss. Finally, fault features are extracted by iteratively refining the estimated periodicity and removing redundant modes. Simulation and experimental results demonstrate that SHNRD can adaptively and accurately decompose single and compound fault patterns under strong noise conditions. Additionally, SHNRD outperforms variational mode decomposition (VMD), fault mode decomposition (FMD), and adaptive maximum second-order cyclostationarity blind deconvolution (ACYCBD). It achieves improvements in rotational frequency suppression of 203%, 162.6%, and 138.7%, respectively.