Model based bearing fault detection using support vector machines

This paper deals with the development of a model based method for bearing fault diagnostics. This method effectively combines the information available in the data and the model for efficient classification of the bearing and the type of defect. A four degrees of freedom nonlinear rigid rotor model is used to simulate the rotor bearing system. Precession of the shaft is measured using proximity probes. The deviation of the measurement from the model is used to classify the system. Typically proximity probe data by itself does not contain enough information for accurate classification. However, when the information from the model is incorporated the combined features provide excellent classification performance. Further the use of a model also enables better classification over varying parameters. A support vector machine is used for classification.