An efficient condition monitoring and fault diagnosis method for bearings under multiple working conditions

Traditional model-based fault diagnosis of rotating machinery established upon the well-known observer design and analysis are limited to the choices of parameters. Meanwhile, signal processing-based methods heavily rely on expert experiences to extract features, hence the usability is heavily limited. Besides, existing researches usually carried out in a single working condition while bearings often work in multiple working conditions in applications. To cope with these issues, a novel fault diagnosis method which combines both signal processing and data-driven techniques is proposed. First, empirical mode decomposition and principal component analysis are employed to separate useful signals from the original non-stationary and nonlinear vibration signals. Then the high dimension and nonlinear features are extracted and kernel principal component analysis and linear discriminant analysis are used to reduce the dimension of features. Finally, the optimized support vector machine is adopted for fault classification. To deal with the few fault sample problem, overlapping sampling is utilized to enhance the data. The proposed methodology is able to conduct fault diagnosis efficiently and precisely in multiple working conditions. Experimental results showed that the prediction accuracy is satisfied even in the case of relatively few faulty samples.