Improved multi-scale entropy and it's application in rolling bearing fault feature extraction

Multi-scale entropy has long been considered as a popular tool for measuring the complexity or detecting the nonlinear dynamic changes of time series, and can be utilized effectively to extract fault features from rolling bearing vibration signals. Nevertheless, with regard to information aliasing, which produces unreliable measurement information, it has been found to consistently exist in the coarse-graining process of multi-scale entropy due to the reduction of data points. In this paper, the adverse information aliasing phenomenon is analyzed, and subsequently two novel nonlinear dynamic methods called local maximum multi-scale entropy and extended multi-scale entropy are proposed. In the proposed methods, the coarse-graining process is endowed with clear physical meaning. Experimental results indicate that the proposed multi-scale analysis methods are suitable for both sample and permutation entropy. Compared with the original methods, the methods recommended in this paper hold better determination, stability and wider analysis scale, while being able to characterize the rolling bearing vibration signal more effectively. (C) 2019 Elsevier Ltd. All rights reserved.