Prediction reliability improvement on long-term creep life for P91 steel using a hybrid method of artificial neural network and CDM model

In present study, a hybrid method had been proposed to improve the low prediction reliability of long-term creep life for P91 steel using a single conventional model. It combined the continuum damage mechanics (CDM) model with the error-trained back-propagation artificial neural network (BP-ANN) model. Firstly, both the conventional CDM model and the Larson-Miller (L-M) parameter method were employed to predict the long-term creep life. Subsequently, the shortterm rupture life data of P91 steel obtained with creep experiments and the related data from the National Institute for Materials Science (NIMS) database and other researchers were utilized to train the hybrid method, respectively. Meanwhile, the extrapolation optimization of the CDM model was carried out with relative errors from the error-trained BP-ANN model. Consequently, the long-term creep life of P91 steel was obtained using three different models. The results demonstrated that, in contrast to the L-M parameter method and single CDM model, the present hybrid method with training data lower than 10,000 h exhibits enhanced prediction reliability of the creep life of P91 steel up to 200,000 h, with a relative error of less than 8 %.