LAUSR

Abstract Planetary gearboxes vibration signals are highly complex due to the intricate kinematics and the amplitude modulation and frequency modulation nature, and their spectra have intricate sidebands, leading to difficulty in planetary gearbox fault identification. To address this issue, multi-domain features are extracted for fault identification from vibration signals to construct feature vectors in time-domain, frequency-domain and through instantaneous amplitude energy analysis. Variational mode decomposition (VMD) is utilized for the merits which are entirely non-recursive, free from pseudo mode and negative frequency and much robust to noise to decompose a signal into a set of mono-components, based on which the instantaneous amplitude energy is calculated. In order to inhibit and eliminate the undesirable features while concentrate the fault identification information to several outperforming features, sparse filtering which is essentially hyper-parameter-free is applied to extract sparse features without considering the selection of k-nearest neighborhood involved in manifold learning. Finally, the proposed method is validated via analyses of planetary gearbox dataset collected in both lab experiments and a wind farm. Based on clustering analysis of multi-domain sparse features extracted, the localized faults are successfully identified.