Fault Intelligent Diagnosis Technology Analysis Of Gearbox Based On SVM

The rotating machinery account for a large proportion in the mechanical equipment, and gearbox is one of the important variable transmission parts of the rotating machinery. Its operating situation is related to the working conditions of the whole machine. Because of the special working environment of the gearbox, its components is susceptible to have some fault. Therefore, there is practical significance by making the gearbox as the subject for the research of mechanical fault diagnosis and condition monitoring.Intelligent fault diagnosis is a growing multidisciplinary which involves a number of areas, although lots of methods and theories have been studied for many years, there is not a mature and complete theory of intelligent fault diagnosis, many problems still needs further exploration and research. The focus of this paper is fault diagnosis of the gearbox by combining business needs.Some typical faults were studied, for example, the diagnosis method and generation mechanism of the gear fault and bearing fault. To study the fault feature extraction, finally, achieving an intelligent fault diagnosis for the gearbox by support vector machine (SVM).As for fault feature extraction, envelope analysis based on spectral kurtosis which is an effective method for fault feature extraction is studied, a pre-whiting process is did for the fault signals using autoregressive (AR) model to enhance impact fault signals of gearbox. Then do fault feature extraction using an improved envelope analysis approach based on complex shifted Morlet wavelet and spectral kurtosis. Finally, the fault feature frequency as the center frequency, the fault feature value which include energy calue, peak value and kurtosis value etc. is calculated of each frequency band in50Hz bandwidth, then they were formed matrix as a data set for SVM.As for fault intelligent diagnosis, genetic algorithm (GA) parameter optimizing method and SVM are studied. With cross validation (CV) method, the accuracy in CV sense as the fitness function value of GA, then using GA to optimize the parameter of SVM. It is effective for avoid overfitting and underfitting occurring and will get an ideal accuracy for test sets finally. Simulation and actual tests results verified the effectiveness of the method.