Creep Life Prediction of CrMo Steel Based on Machine Learning

Jiaqi Wang, Yongzhe Fa, Xinghua Yu*

Beijing Institute of Technology , Beijing 100081 , China

ABSTRACT: CrMo series alloy steel is the main material of most modern power plants and boilers. High temperature and high pressure environment greatly increases the tendency of steel creep fracture and reduces the creep life. Therefore, it is of great practical significance to reasonably predict the creep life of steel. Due to the long time from beginning of creep to creep fracture of materials, it is limited to predict the quantitative relationship between properties of metal materials and their composition and processing parameters by using traditional methods based on repeated tests and computational simulation. Material genetic engineering uses data-driven and machine learning technology to analyze material data, which can effectively avoid the shortcomings of traditional methods such as time-consuming, accelerate material design and research, and has a wide application prospect. Based on a large number of existing material data, machine learning method including many regression models is used for data mining and prediction, which provides a new possibility for creep life prediction of CrMo series alloy steels. Creep life of materials is usually affected by many factors , so it is very important to study the influence of multi parameters on creep life. Based on the steel creep data set developed by phase change group of Cambridge University, principal component analysis (PCA) and clustering methods were used to analyze the original creep data.At the same time, nine different machine learning regression models were used to predict the creep life of CrMo alloy steels. Finally, the relationship between creep life of CrMo series alloy steels and many descriptors including alloy composition and heat treatment conditions was comprehensively studied. We transformed creep life into time-temperature parameters, including Larson-Miller parameter, Manson-Haferd parameter and Manson-Succop parameter. These parameters were used to replace original creep life for prediction. Results show that the R² value of model trained by time-temperature parameters is larger, the MAE_c and RMSE_c values are smaller, and the prediction accuracy is higher. Among all the models, random forest model has the best prediction effect on creep life, and the model trained by Manson-Succop parameter has better stability and accuracy. Among all the 20 input characteristics, pressure and Nb content are the main factors affecting creep life of CrMo alloy steels.

Keywords: CrMo steel; Creep life; Machine learning; Performance prediction