A blind deconvolution separation of multiple sources, with application to bearing diagnostics

Abstract

A method is presented, geared to separating signals from different sources which are convoluted and mixed by the mechanical systems before being measured. The method is based on an automatically operating blind deconvolution separation method, with Kurtosis of the separated signals as the measure to be maximised. The application described involves bearings diagnostics, whereas with many classical diagnostic methods, Kurtosis is traditionally one of the accepted criteria for fault detection. The method is tested on simulation and experimental cases. Results show that separation is possible even when measurements are distanced from the vibration exciting sources of the faulty bearing. Furthermore, the method eliminates the effect of structural resonances, which often causes severe problems in classical diagnostic methods.