Blind source separation of composite bearing vibration signals with low-rank and sparse decomposition

Fault diagnosis is pivotal for health monitoring of rotating machinery. On practical engineering occasions, collected signals are usually from multi-sources. Moreover, the complex transmission path between multi-source and sensors further complicates the situation. So existing conventional methods for blind separation and detection could not meet the requirement in complex operating conditions. Inspired by the success of robust principal component analysis in sound signal processing, we proposed a novel strategy that explores the rank and sparsity features of signals for blind separation and detection of composite faults. The main methodological contribution of this paper is to consider the problem of bearing faults from the perspective of signal's rank and sparsity, which are firstly explored in time-frequency domain. And their diversities are utilized to carry out further separation. Experiment shows that the proposed strategy can achieve target detection and multi-source separation, making it more effective to separate and detect bearing composite faults.