Feature extraction method for roller bearing based on Dempster-Shafer evidence

Abstract

The fault features of roller bearing vibration signal are usually immerged in heavy background noise and difficult to be extracted. Especially for the diagnosis of weak fault, it is even harder. To extract the weak fault features of bearings effectively, a new method which used complementary ensemble empirical mode decomposition (CEEMD) and Dempster-Shafer (DS) evidence theory is proposed. Firstly, the noise reduction of original signal based on CEEMD is carried out. The vibration signal of roller bearing is decomposed into a series of intrinsic mode functions (IMFs) by CEEMD. The cross correlation coefficients between the original signal and each IMF are calculated to select effective IMF. Secondly, the data-based fusion of effective IMFs is carried on by DS fusion rules. The fault feature components and interference components in the vibration signal can be considered as two incompatible propositions of DS “recognition framework”. The effective IMFs represent the evidences supporting the fault features. Fusing the effective IMFs to integrate the local incomplete information of single IMF and enhance the weak fault features of bearings. Finally, the fault features can be extracted after Hilbert envelope demodulation. The effectiveness of the presented method is validated by simulation and experimental signal. And the results indicate that the proposed method is available for detecting the bearing faults and able to diagnose the weak fault at an early stage.