53 SVM-Based Prognosis of Machine Health Condition

Machine condition prognosis is a method to predicting machine condition deterioration, based on which condition based maintenance can be effectively scheduled to prevent unexpected machine breakdown, injuries of human, and costs due to the loss of productivity. The noise present in the measured condition variables may adversely affect the prognostic results. Reported work to address this problem includes removing the effects of noise from the observations of condition variables before prediction and modeling actual observations in a statistical way, which is, however, usually not user-friendly. In this work, support vector machine (SVM) is used directly for machine condition prognosis. As SVM is a parametric method, it has potential to achieve good denosing and predicting results simultaneously by appropriately adjusting its model parameters. An optimization model is developed for selecting the parameters, which is solved by adopting genetic algorithms.