Sparse Representation Based on MCKD and Periodic Dictionary for Bearing Fault Diagnosis

The sparse representation (SR) is widely used in bearing fault diagnosis for its high accuracy in fault signal restoration. However, when the original signal contains intense background noises and interference pulses, the performance of SR often turns out to be poor. To solve the existing problems, this article proposes an SR method based on the maximum correlated kurtosis deconvolution and periodic dictionary (MCKD-PDSR) for the bearing fault diagnosis. In MCKD-PDSR, the dictionary of SR is constructed based on the filtered signal of MCKD, so that the accuracy of the dictionary is improved. Furthermore, a periodic dictionary and its construction process are proposed, which enables SR to identify periodic fault signals and filter out the noise and interference components. The simulation results show that MCKD-PDSR has higher signal recovery accuracy than the traditional SR. The feasibility of the proposed method is verified by the data of the bearing life acceleration experiment. The results indicate that MCKD-PDSR is capable of extracting the fault signal and is useful for the fault diagnosis in a complex environment.