Early Failure Detection of Bearing Based on Probabilistic Matching Pursuit

A signal adaptive time-frequency decomposition to extend matching pursuit algorithm-probabilistic matching pursuit is studied. This method is firstly introduced into early fault diagnosis of rolling bearing. It adopts Gaussian wavelet to construct the time-frequency atoms dictionary. Applying probabilistic matching pursuit, we can accurately extract the early defect features by computing coherence between vibration signals of rolling bearing with inner or outer race fault and the dictionary elements. Comparing with matching pursuit, probabilistic matching pursuit represents vibration signal features with very fine resolution and sparsity in the time-frequency domain through computer simulations. Simultaneously, experiment results confirm that probabilistic matching pursuit is quite effective to facilitate the early diagnosis and identification of rolling bearing faults.