Feature extraction for bearing fault diagnosis using composite multiscale entropy

Abstract

Multiscale entropy (MSE) is a popular algorithm to measure the complexity of a time series for multiple scales. However, the conventional MSE algorithm yields imprecise estimation of entropy for a time series with large time scale factors. In this paper, a composite multiscale entropy (CMSE) method is proposed to overcome this drawback. In the CMSE algorithm, with scale factors of τ, we calculate the sample entropies (SampEns) of all coarse-grained series and then define the mean of τ SampEns as the entropy values. This proposed algorithm is then applied to two different kinds of simulated noise signals and a set of real vibration data. These results demonstrate that the proposed CMSE provides more precise entropy calculation than the convectional MSE. Furthermore, as a feature extractor for a bearing faulty signal, CMSE provides a higher distinguishability, compared with MSE.