Abstract: Aiming at the features extraction of rolling bearing vibration signals and the rapid and effective diagnosis of fault types, a rolling bearing fault diagnosis model based on EMD-AR spectrum and GA-BP neural network was proposed. Taking the bearing vibration acceleration data from the Bearing Data Center of Case Western Reserve University (CWRU) as an example, the theoretical analysis and experimental testing of various states of rolling bearings were carried out. First, the collected bearing vibration signal was decomposed by empirical mode decomposition (EMD) to obtain the intrinsic modal function components of different orders, and then these components were analyzed by auto regressive (AR). The fault feature was extracted and the fault feature vector matrix was formed by the parameters of the AR model and the variance of the residual. Finally, the fault feature was used as the input data and output data of the genetic algorithm to optimize the BP neural network for training and testing.The research results show that the rolling bearing fault diagnosis method based on EMD-AR spectrum and GA-BP can effectively identify different types of fault characteristics. Comparing with the traditional BP neural network, GA-BP neural network has a higher diagnosis efficiency, and the accuracy rate can reach 95%.

Key words:  rolling bearing; empirical modal decomposition(EMD); auto-regressive (AR); GA-BP neural network; fault diagnosis