Bearing initial fault feature extraction via sparse representation based on dictionary learning

As initial fault occurs in rolling bearings of low-speed and heavy-duty machinery, the impulse component, reflecting the fault feature in vibration signals is difficult to extract because it is relatively weak and easily corrupted by strong background noise. The authors attempted to extract the impulse component from a vibration signal with the sparse representation method. However, it is difficult to construct an accurate dictionary that matches the impulse component since operation conditions of bearing are not stable. Hence, a method of extracting the initial fault feature, which is based on dictionary learning, was proposed in this research. Firstly, an adaptive dictionary was obtained by the developed K-SVD dictionary-learning algorithm. Then, Orthogonal Matching Pursuit (OMP) algorithm was utilized for sparse decomposition of the vibration signal, and all kurtosis values of approximation signal of iterations were calculated. Finally, the corresponding approximation signal of maximal kurtosis value was reconstructed and analyzed with the envelope spectrum to diagnose the fault type. The test results of simulated data and bearing vibration signals demonstrate that the proposed method, which can extract the feature component more accurately than other methods, meets the demand of real-time bearing condition monitoring. © 2016, Editorial Office of Journal of Vibration and Shock. All right reserved.