ZHANG Ying1, MA Bo1, ZHANG Ming1, YANG Lu wei2, YANG Jun ling2

 

(1. Diagnosis and Self Recovery Engineering Research Center, Beijing University of Chemical Technology,

 

Beijing 100029, China; 2. Key Laboratory of Cryogenics, TIPC, CAS,Beijing 100190, China)

 

 

Abstract: 

Aiming at the problem that fault signal of rolling bearing was non stationary,aiming at which, how to extract feature of non stationary signal was studied and the methods based on empirical mode decomposition (EMD) and principal component analysis (PCA) was put forward .Signal was decomposed into several IMFs by means of EMD. The energy of every intrinsic mode function (IMF) was calculated,and some IMFs with greater energy was chosen. Every IMF was divided into several segments according to the range the frequency,and energy of every segment was calculated as the feature value. PCA was used to reduce the dimensions of feature value . Accumulative contribution of feature value was calculated and the first few feature values whose accumulative contribution rate reached 80% was chosen as the final fature value . The results indicate that the method is valid for the feature extraction of rolling bearing , and dimension reduction of high dimensional feture.

 

Key words: empirical mode decomposition (EMD); principal component analysis (PCA); rolling bearing; feature extration; nonstationary signal