Combining refined composite multiscale fuzzy entropy (RCMFE) and support vector machine (SVM) with particle swarm optimization (PSO) for diagnosing roller bearing faults is proposed in this paper. Compared with refined composite multiscale sample entropy (RCMSE) and multiscale fuzzy entropy (MFE), the smoothness of RCMFE is superior to that of those models. The corresponding comparison of smoothness and analysis of validity through decomposition accuracy are considered in the numerical experiments by considering the white and 1/f noise signals. Then RCMFE, RCMSE and MFE are developed to affect extraction by using different roller bearing vibration signals. Then the extracted RCMFE, RCMSE and MFE eigenvectors are regarded as the input of the PSO-SVM to diagnose the roller bearing fault. Finally, the results show that the smoothness of RCMFE is superior to that of RCMSE and MFE. Meanwhile, the fault classification accuracy is higher than that of RCMSE and MFE.
