A study of creep rupture life prediction for P91 steel with machine learning method: Model selection and sensitivity analysis

Abstract

The present work dealt with the long-term creep rupture life of P91 steel using the machine learning method. In order to achieve the reliability evaluation, different machine learning algorithms were selected to train the test data of P91 steel, respectively. In particularly, a large model utilizing the creep data of different types of P91 steel was proposed to improve the relatively large error in the direct prediction. The results revealed that in contrast to other models, the predicted long-term creep rupture life of P91 steel with a relative error of less than 4 % was obtained with the multi-layer perceptron algorithm. Moreover, sensitivity analyses suggested that the predicted creep rupture life was strongly dependent on input factors, such as the chemical composition and heat-treatment condition. Additionally, the proposed machine learning framework not only provided the reliable creep rupture life prediction, but also offered a potential tool for material engineers to design and produce the high-performance P91 steel.