Abstract: In order to effectively improve the reliability and accuracy of rolling bearing fault diagnosis, in view of the advantages of refined composite multiscale reverse fluctuation dispersion entropy (RCMRFDE) method in describing the complexity of nonlinear sequences and feature extraction, a comprehensive rolling bearing fault detection method combining RCMRFDE and extreme learning machine (ELM) was proposed (this method includes health detection and fault classification).Firstly, according to advantage of the significant difference between the health and fault vibration signal complexity of rolling bearing, RCMRFDE was used to detect the health state of rolling bearing in advance and screen out the healthy bearing. Then, RCMRFDE was used to extract the fault features of the remaining fault bearings, and the fault type was intelligently recognized based on the extreme learning machine(ELM).Finally, based on two kinds of published rolling bearing fault data, the RCMRFDE+ELM method was tested with other five fault diagnosis methods, and the obtained results were compared,to verify the detection accuracy, classification accuracy, efficiency and reliability of the new method.The research results show that the proposed method can accurately detect whether there is a fault in the rolling bearing, and the average recognition accuracy of the fault rolling bearing was 99.96% and 99.67% respectively, which were higher than other methods. It also provides a detection threshold setting scheme and a fault diagnosis idea for establishing the health monitoring model of rolling bearings.
Key words:  bearing fault diagnosis; fault feature extraction; bearing health detection; fault classification; refined composite multi-scale reverse fluctuation dispersion entropy(RCMRFDE); extreme learning machine(ELM); comprehensive fault detection