Enhancing the ability of Ensemble Empirical Mode Decomposition in machine fault diagnosis

Abstract

Empirical Mode Decomposition (EMD) is an adaptive time-frequency analysis method that has been widely employing for machinery fault diagnosis. EMD is famous in revealing instantaneous change of frequency or time from non-linear sensory signal so that the occurrence of anomalous signal can be accurately detected. However, its shortcomings include mode mixing and end effects that often appear in its decomposed bands. These problems decrease the accuracy, particularly in vibration-based fault diagnosis. Recently, many researchers have proposed various improved methods, which include the famous Ensemble EMD (EEMD), to solve the problem of mode mixing. Its purpose is to introduce controlled amount of white noise to the original EMD. After adding known white noise into the raw signal, the signal in the band will have a uniformly distributed reference scale which forces the EEMD to exhaust all possible solutions in the sifting process for minimizing mode mixing effect. Even though EEMD becomes popular, the proper settings for the number of ensemble and the amplitude of white noise that should be added are still not formally prescribed. This paper discusses the influence of parameters setting on the results of reducing mode mixing problem. Tests were done using both simulated and real machine signals. Their results provide a guideline on setting the parameters properly so that the ability of EEMD on machine fault diagnosis can be significantly enhanced.