Research on Application of Supported Vector Machine to Creep-Fatigue Lifetime Prediction of P91 Steel

Abstract

Creep-fatigue tests were carried out to study creep-fatigue property of P91 steel in many references. There are two kinds of loading control methods. One is controlled by stress, and the other is controlled by strain. In this study the creep-fatigue test data loading controlled by stress are summarized. Based on these data Supported Vector Machine(SVM) method is used to build up the models to predict the creep-fatigue lifetime of P91 steel. The results show SVM method can be used to predict the creep-fatigue life of P91 steel, and the value of insensitive factor is very important. When the value of insensitive factor is small and the training error of the model is small too, but the prediction ability decreases. On the contrary when insensitive factor is increased, the training error will become large, and the prediction ability will increase. So during the training, mean square error should not be regarded as the unique goal function.