Refined composite moving average fluctuation dispersion entropy and its application on rolling bearing fault diagnosis

Abstract

Vibration signal analysis based on multiscale entropy is one of the important means to realize rotating machinery fault diagnosis. However, the length of the time series will be shortened during the coarse-graining process with the increase of the scale factor, which makes the calculated entropy values unstable. This inherent drawback of the coarse-graining method limits its application in fault feature extraction. This paper presents a novel feature extraction method for vibration signals called refined composite moving average fluctuation dispersion entropy (RCMAFDE). It is verified by simulation experiments that RCMAFDE has high stability of entropy values under different time series lengths as well as different disturbances. The RCMAFDE was applied to the fault diagnosis of rolling bearings, and a new fault diagnosis method of rolling bearings was proposed by combining RCMAFDE and kernel extreme learning machine (KELM) optimized by the chaos sparrow search optimization algorithm (CSSOA). First, the vibration signal is preprocessed to form a sample set, and then, the fault feature vector is calculated by RCMAFDE. Finally, the feature vector set is input into the CSSOA-KELM model for training and testing, and the fault diagnosis result is output. To demonstrate the effectiveness and feasibility of the fault diagnosis method, two publicly available datasets and a self-collected dataset are used for experimental validation. The experimental results show that the proposed fault diagnosis method can extract the nonlinear dynamic complexity information of vibration signals more effectively compared with the comparison methods and obtain the highest fault identification accuracy under different datasets.