A New Intelligent Method for Bearing Fault Detection Based on Co-integration Concept and Selecting the Optimal Feature Set at Time-Varying Speed Conditions

In this paper, a new hybrid intelligent method is suggested for the bearing fault detection at time – varying speed conditions. The vibration signals have been collected for two states as healthy bearing and defected inner race under variable speeds. In this study, the ensemble empirical mode decomposition (EEMD) technique and Johanson trace method are utilized for extracting the co-integration relationships from the vibration data. Then, the feature matrix corresponding to the co-integration relationships is calculated using the wavelet packet decomposition (WPD) method, and the time -domain statistical features. In the next stage,the compensation distance evaluation technique (CDET) has been used to determine the preselected feature subsets. The preselected features are utilized as input data of the support vector machine (SVM) to predict the bearing state. Finally, The optimal SVM parameters and the optimal feature subsets are determined using the binary particle swarm optimization (BPSO) algorithm. The obtained results demonstrate that the optimal features are well able to differentiate between different bearing states at time-varying speeds. Comparing the results of this article with other fault detection methods indicates the ability of the proposed method.