Sensible multiscale symbol dynamic entropy for fault diagnosis of bearing

Abstract

Due to the complex service conditions of rolling bearings, vibration signals arising therefrom exhibit non-linear characteristics, which means that single-scale feature extraction techniques cannot extract fault features. Multiscale symbolic dynamic entropy (MSDE) is a new technique that has recently emerged and been applied to fault diagnosis in machinery. However, MSDE has limitations such as its poor stability, large errors, and even loss of information. To this end, a novel sensible multiscale symbol dynamic entropy (SMSDE) method was proposed. For SMSDE, the signal was first decomposed using empirical mode decomposition, and then the useful intrinsic mode functions were selected for reconstruction to decrease noise. Secondly, the slippage-averaging multiscale approach was designed to coarse-grain the signal, which considers the connection of data before and after the breakpoint, thus reducing the error. The method can not only decrease noise, but also avoid the loss of key information, thereby extracting sensitive feature information. The results with multiple synthesized signals show that the proposed method is more robust than the other eight entropy methods. Furthermore, the real bearing signals of the three cases indicate that compared with other advanced entropy methods, SMSDE can better distinguish the various states of the bearing.