|Ph.D Student||Peled Ran|
|Subject||Blind Source Separation (BSS) of Vibrations for Fault|
Detection in Rotating Systems
|Department||Department of Mechanical Engineering||Supervisors||Professor Emeritus Simon Braun|
|Professor Miriam Zacksenhouse|
This work deals with blind separation of sources when acquiring signals from mechanical systems, when no a priori information is used.
When several exciting sources exist, most of them cause a response at the monitored point. The unknown propagation properties result in a complicated mixture of signals modified by structural properties.
In this work our method is applied to the case of roller bearing monitoring, when multiple bearings excite vibrations. The analyzed signals are of the type of periodic impacts, typical of bearing faults.
Signals, propagating in a mechanical structure, are modified by its dynamic properties. Our separation algorithm is based on approximate inverse filters, which then separated sources.
Tested were simulated and experimental data. The experimental case is one of a shaft supported by three bearings, with a localized fault in one of them, with measurements near and far from the faulty bearing, and attempting to separate the contribution of the faulty bearing.
We show the feasabilty of identification sources. The impacting signals were identified more clearly than in the raw measurements, with the effect of the dynamic structure decreased. We show that using our method, it is possible to limit the analysis to the region of low frequencies, below the resonances, after the separation process.
The separation improved monitoring even for measurements taken close to the fault. For difficult situations where close measurements were not possible, the use of our method improved the source analysis significantly.