The EMD and classifier ensemble-based ball bearing fault diagnosis method

To address the problem regarding the nontranparency of ball bearing fault diagnosis process, a study is made by taking the following steps: construction of the rule-based classifier ensembles based on the six time-domain features and five frequency-domain features extracted through empirical modal decomposition and imitation of the thinking and inference process of technical personnel; singling out the optimum base classifiers through screening according to diversity indices; determination of the reduction and diagnosis rules of the candidate base classifiers using genetic algorithm; and formation of the classifier ensembles by using the weighted voting strategy. Practice shows with the use of the method for identifying the failure of the inner ring, outer ring and balls of a normal ball bearing running at different speeds, the correct identification rate is up to 90%. Unlike the black-box models, the diagnosis process can be proceeded in a way similar to the reasoning process of a technician, featuring a good interpretability-a method which is more likely to be accepted by technical personnel working on site.