Low-rank and sparse model: A new perspective for rolling element bearing diagnosis

Rolling element bearings are one of the most essential yet vulnerable components in railway systems. Due to harsh working conditions, its health condition degrades over time, or worse still, the incipient fault signature is easily overshadowed by strong interfering noise. In recent years, sparse representations of the vibration signal have attracted more and more attention in the scientific community. The benefits of sparsity-based models have been fruitfully explored and, in brief, most of the research matches the fault signature of vibration signal to some degree which addresses specified problems rather than general ones. The state-of-the-art in sparsity, as it applied to machinery fault diagnosis, still faced some challenges. Of particular importance is dealing with the incipient fault, especially in the case of poor signal-to-noise ratio (SNR). The aim of this communication is to fill in the gaps by introducing the low-rank and sparse model (LRSM). First, recent sparsity-based models are succinctly reviewed with their pros and cons. Second, a new model based on the low-rank and sparse constrains is proposed for capturing the repetitive transients embedded in heavy background noise which jeopardize their symptoms in practical applications. Eventually, its effectiveness is investigated on synthetic signals in the case of low SNR=-6dB.