Independent component analysis based source number estimation and its comparison for mechanical systems

Abstract

It has been challenging to correctly separate the mixed signals into source components when the source number is not known a priori. In this paper, we propose a novel source number estimation based on independent component analysis (ICA) and clustering evaluation analysis. We investigate and benchmark three information based source number estimations: Akaike information criterion (AIC), minimum description length (MDL) and improved Bayesian information criterion (IBIC). All the above methods are comparatively studied in both numerical and experimental case studies with typical mechanical signals. The results demonstrate that the proposed. ICA based source number estimation with nonlinear dissimilarity measures performs more stable and robust than the information based ones for mechanical systems.