Abstract:In practice, it is very difficult to obtain the rubbing fault samples, but the non-rubbing normal samples are very rich, therefore, in this paper the one-class support vector machine(SVM) is introduced to recognize the rubbing fault, which can obtain the recognition border of rubbing fault through learning from only the a lot of normal samples. Because the rotor fault signal spectrum features are very redundant, a new feature extraction method based on the Primary Component Analysis (PCA) is put forward, firstly, the rotor fault signal frequency spectrum is normalized; secondly, the spectra data of a lot of samples is carry out Primary Component Analysis, and the lower dimensions features are extracted according to different energy preserving rate. Finally, the new approach is verified through some diagnosis experiments.