Independent component analysis as applied to vibration source separation and fault diagnosis

Abstract

In health monitoring of complex mechanical systems such as aircraft engines there are many components whose diagnosis is of great interest for the industry. A conventional way to monitor these components is to collect vibration signals using accelerometers placed in their closest vicinity. However, due to some restrictions such as inaccessibility, it is not always practical to place the accelerometers as such. In many cases, pre-installed instrumentations are used, which are usually inadequate and placed on the carcass of the structure. Nevertheless, even if the accelerometers are positioned very close to the components, they would collect signals not just from one specific component but from other components as well. In this study, we sought to employ frequency-based independent component analysis (ICA) to recover the signals produced by components within a single complex system. In such a case, differences between “blind source separation” and vibration source separation are discussed. A new workaround for the permutation ambiguity encountered in the implication of ICA is proposed. Finally, in order to demonstrate the applicability of the new proposed approach, experimental results carried out on a test bed are presented.