Initial Fault Feature Extraction of Bearing Based on Sparse Representation

Rolling bearings of low-speed and heavy-duty machinery work under huge load, therefore they are easily gotten inner or outer race faults. In initial fault stage, the impulse component, reflecting the fault feature in vibration signal, is difficult to extract for it is relatively weak and easily corrupted by strong background noise. A fault feature extraction method based on sparse representation was proposed to accurately diagnose the initial fault of bearing. The method utilized K-SVD dictionary training algorithm for constructing an accurate dictionary to match the impulse component and overcome the problem of parameter dictionary lack of adaptability. In sparse coding, batch orthogonal matching pursuit (Batch-OMP) algorithm was employed to sparse-decompose the vibration signal, and the kurtosis maximum principle of approximation signal was the end condition of decomposition, which determined the decomposition times adaptively. Finally, the feature component was reconstructed and its envelope spectrum was analyzed to diagnose the fault type. The fault feature was extracted by the proposed method from simulate and bearing vibration signals. The results show that the method can extract the impulse components accurately, which demonstrates its effectiveness and practicability.