Feature extraction of gear and bearing compound faults based on vibration signal sparse decomposition

Abstract

Compound faults of gear and bearing in a gearbox tend to couple features both of distributed and localized defects. The vibration signal shows overlapped modulation phenomena, which cause most traditional wave-filtering-based diagnosis methods invalid. A sparse-decomposition-based method is proposed to decouple overlapped modulation signals and extract features of gear and bearing compound faults. Two kinds of dictionaries respectively consisted of steady harmonic atoms and transient impact atoms are designed to match features of compound faults. Atom parameters are self-adaptively identified from the spectrum information, and identification precisions are improved by the techniques of discrete spectrum correction and correlation filtering. Fault features respectively related to the distributed and localized defects are successively extracted by a novel piecewise matching pursuit algorithm. Lastly, the compound impact features of gear and bearing localized defects are further separated according to the impact period differences in time domain. Both simulation analyses and experimental tests verified the proposed method’s effectiveness on the diagnosis of gear and bearing compound faults.