A condition based predictive maintenance system will have the following functions: analyzing collected vibration signals and identifying the components that have deteriorated significantly, predicting the degradation of these components, and making an appropriate maintenance plan for minimizing total equipment operation cost. In this thesis, Support Vector Machines (SVM) regression is used for prediction of machine's degradation and is studied in depth. The selection of SVM model parameters is investigated based on current problems that have arisen in the industrial application of SVM regression. A new rule is proposed for selection of the error zone value, one of the SVM model parameters. The proposed rule is also compared with CMa's method and the results show that applying the new rule, SVM regression can provide better prediction than CMa's method. |