Rolling bearing diagnostics by means of EMD-based independent component analysis on vibration and acoustic data

Independent Component Analysis (ICA) is used for recovering the various independent sources exciting a system and to separate their contributions. In this paper, ICA is applied to vibrational and acoustic data measured on undamaged and damaged rolling bearings, where those data contains information about the vibration frequencies related to the defect, if present, to the rotation of the bearing, to its dynamic behaviour (resonance frequencies) and to random noise. The aim of the method is to separate those different contributions and to enhance only the one related to the vibration frequency associated to the defect in order to diagnose its presence. To improve the power of the method to identify the contribution of the bearing fault characteristic frequency, a pre-processing step is introduced based on Empirical Mode Decomposition (EMD), allowing to reconstruct multiple sets of time series from a single sensor data, called multiple intrinsic mode functions, which are then given as input to the ICA.