A novel fault diagnosis method based on EMD, cyclostationary, SK and TPTSR

A novel method based on empirical model decomposition (EMD), cyclostationary, spectral kurtosis (SK) and two-phase test sample sparse representation (TPTSR), called ECK-TPTSR is proposed for fault diagnosis in this paper. In the ECK-TPTSR method, the vibration signal is decomposed into several components by EMD. Then each component can be modelled as cyclostationary for noise reduction. Next, the proposed method computes the kurtosis of the unbiased autocorrelation on the squared envelope of each component, and extracts the component with the highest kurtosis. Finally, the extracted component is used to construct training samples and test samples, which are input into the TPTSR classifier to fulfill fault classification accurately. Moreover, the experimental results indicate that the ECK-TPTSR method can effectively achieve fault diagnosis of motor bearing and obtain higher classification accuracy.