Dense multi-scale entropy and it’s application in mechanical fault diagnosis

Abstract

Multi-scale entropy (MSE) is a widely recognized feature extraction approach to mechanical fault diagnosis, for it can effectively estimate the complexity of nonlinear time series. For MSE algorithm, due to the sensitivity of entropy estimation on series length, the scale factors are often required to be limited to a small range. Nevertheless, in the existing MSE methods, the scale factors can only be set to positive integers with a fixed minimum step size, which may result in insufficient analysis precision and cannot provide high-quality feature vectors with sufficient eigenvalues for intelligent diagnosis in the limited scale range. In view of the above defects, this paper subdivides the scale factors and proposes dense multi-scale entropy. In the suggested method, the number of data points in the raw sequence is expanded on the premise of guaranteeing the characteristics of the original series. Based on this, the timescale of the original series is refined and more intensive scale factors with higher precision can be provided. The superiority of the method developed in this work is verified by using CWRU bearing and reciprocating compressor gas valve fault data sets, and the results indicate that the method of this paper can provide more precise analysis scale and feature vectors with higher quality for intelligent diagnosis.