Novel cyclostationarity-based blind source separation algorithm using second order statistical properties: Theory and application to the bearing defect diagnosis

Among signal processing techniques, blind source separation (BSS) and the underlying mathematical tool of independent component analysis (ICA) are of continuously growing interest in the scientific community of various research domains. Vibration analysis is a potential application field of this quite recent technique. Actually, BSS methods aim to retrieve unknown source signals from a set of their observations coming to a matrix of sensors, without necessarily having any prior knowledge about the sources. In monitoring and diagnosis purposes, bearing defects constitute a problem for manufacturers who aim at predicting those faults as well as potential engines breakdowns. These defects may be the unknown sources one wants to estimate from a set of recorded signals by a matrix of accelerometers placed close to the rotating machine. It has been shown that these vibration signals are wide-sense cyclostationary [ [11] R.B.Randall, J. Antoni, S. Chobsaard, The relationship between spectral correlation and envelope analysis in the diagnostics of bearing faults and other cyclostationary machine signals, Mechanical Systems and Signal Processing 15 (5) (2001) 945–962]. The new algorithm of BSS proposed in this work is based, precisely, on that property. Second-order statistics of such processes led us to a new separation criterion for blind source separation. The theoretical results of this study, simulation and experimental analysis are presented in here. Perspectives for future research conclude this paper.