Blind Enhancement of Harmonically Related Signals by Maximizing Skewness

Sammanfattning: Rolling element bearings are used in rotating machinery in various industry branches. Their health status must be monitored continuously in order to establish proper operational conditions in a production process. Numerous approaches, which can be investigated under the subject of ``Condition Based Maintenance", have been studied within mechanical engineering and signal processing to be able to detect and classify possible faults on rolling bearings.Periodic impulsive signals can emerge from defected bearings within rotating machinery. As the signal is distorted by an unknown transfer function, noise and severe interference, the challenge becomes to reduce these effects as much as possible to extract valuable and reliable information about the rolling bearings' health status. Without any observation of the source signal, a scale-invariant higher order moment, skewness, can be used as a tool to characterize statistical properties to enhance the desired signal. It is the impulsiveness, thus asymmetry of the signal that will be promoted. To assess the performance of skewness, a signal model that consists of harmonically related sinusoids representing an impulsive source is built. Depending on such a model, surface characteristics of skewness are investigated. In relation to harmonic content, the ability of skewness in discovering such harmonic relation is studied. It has been observed that the optimization process converges to a setting where all harmonics are preserved, while any component that does not possess such a harmonic relation is suppressed. In the case of multiple mutually inharmonic source signals with harmonic support, it is shown that skewness maximization results in a setting where only the harmonic set with highest skewness remains. Finally, experimental examples are provided to support theoretical findings.