Approach to finding material with increased long-term strength using mechanical creep model, machine learning and optimization algorithm

Abstract

One of the key problems of automation and energy is the search for advanced materials with increased long-term strength. In this paper, an approach is proposed that combines statistical methods with a physical analytical dependence as applied to finding compositions with increased creep rupture stress in the class of low-alloy steels. The paper uses the Arrhenius creep model to transform data that preserves its quality. When optimizing the composition and selecting the best candidates, an ensemble of machine-learning models is used to increase the generalization ability of the method. Using the approach a new composition is obtained with predicted creep rupture stress increased by 9 % compared to the largest in the dataset.