Harmonic-percussive source separation for fault detection in ball bearings characterized by cyclic polynomial motion law

Bearings are one of the most investigated components leading to failures in rotating machines. Much attention has been paid to the condition monitoring of machines in non-stationary conditions, nevertheless several working conditions have not been adequately studied yet. For instance, the brushless motors used as electric cams are characterized by cyclic polynomial motion laws and motion inversion of the shaft. This paper presents a new approach based on the harmonic-percussive source separation (HPSS) which separate vibration signal into three components, namely harmonic component, percussive component and residual component. The spectral entropy (SE) of the harmonic and percussive signals is devised as a measure of the informative content carried by each component. A prevalence of the SE in the percussive component is attributable to a faulty condition of the bearing, whereas a prevalence in the SE of the harmonic component suggests a healthy condition of the bearing.