Dissimilarity Measures for ICA-Based Source Number Estimation

Abstract

Most of blind source separation problems are carried out with a priori knowledge of the source numbers. However, for source separation-based machinery condition monitoring and fault diagnosis, it is a challenge work to determine the number of sources for a well source separation due to complex structures and nonlinear mixing mode. Therefore, source number estimation is a necessary and important procedure prior to source separation and further diagnosis work. In this paper, we focus on a novel source number estimation method based on independent component analysis (ICA) and clustering evaluation analysis, and investigate the performances of different dissimilarity measures of ICA-based source number estimations with typical mechanical vibration signals. Our work contributes to find an effective solution of source number estimation for source separation-based machinery condition monitoring and fault diagnosis.