采用非相关字典学习的滚动轴承故障诊断方法

To identify the weak fault features of bearing vibration signals at the early stage, incoherent dictionary learning methods are utilized to extract impact features of the signals and diagnose the fault. The dictionary construction is a key step that determines the performance of the sparse representation algorithm. The redundant dictionaries constructed by the traditional dictionary learning methods, however, have a great correlation between atoms. They are not enough to represent different structural characteristics of the signals and are not conducive to the accurate sparse reconstruction of the signals. Therefore, the incoherent dictionary learning method (INK-SVD) is adopted to reconstruct the adaptive dictionaries with low correlation. The low correlation dictionary is learned adaptively in the samples of noisy vibration by the INK-SVD, and the impact fault features are extracted accurately with sparse representation algorithm, so as to achieve more precise bearing fault diagnosis. In terms of the simulation and experiment analysis, the proposed method is able to obtain more accurate sparse coefficients compared with classical dictionary learning methods. Therefore, the envelope demodulation spectrum of the reconstructed signal is more conducive to fault feature identification, which verifies the effectiveness of the method.