Intelligent fault diagnosis method for rolling bearings based on EMD and MLEM2

To solve the problems of automatic fault diagnosis of rotating machinery, an intelligent method based on EMD and MLEM2 was presented. EMD was used to preprocess the original vibration signal, after that, six time-domain characteristic indices and five frequency-domain indices were calculated on the most appropriate IMF to form the dimensionless fault eigenvector of rolling bearings. According to the characteristic vector, fault decision table could be acquired by the data collected from the running machine. The MLEM2 algorithm was then applied to mine diagnostic rules from the data table. By these rules and an improved rule matching strategy, the final fault classification was carried out. EMD could discover the fault essence of the signal, and enhance the signal-to-noise rate of the selected IMF, while MLEM2 algorithm could be operated without attribute discretization, so the result rules were more complete and accurate. It was proved by the experiment of SKF6203 rolling bearings that the accuracy of this method reached 93.75%. It works like an expert system with the ability of acquiring knowledge itself, and does not need any artificial interference once the initialization is made. It is a valid method for intelligent fault diagnosis of rolling bearings.