Insights into irradiation creep coefficient in nuclear graphite from machine learning

Abstract

Understanding irradiation induced creep in nuclear graphite is critical for the service life extension of current reactor fleet and the technological advancement of next generation nuclear reactors. Nevertheless, qualifying a new graphite grade with respect to irradiation creep requires years of testing and expensive facilities for experiments. Here for the first time, we applied machine learning (ML) algorithms to investigate the irradiation creep coefficient in the secondary stage of graphite creep in hope of gaining new insights and expediting the qualification process. Four ML models were trained on a small dataset with temperature and materials properties as input. The gradient boosting regression model exhibits the best predicting performance. The ML models indicate that temperature and Young's modulus are the most important parameters in the determination of creep coefficients while the rest properties have much weaker impact. These findings align with previous theories and corroborate a creep mechanism governed by dislocation climb, demonstrating the potential of ML in improving the workflow of graphite qualification for advanced reactors.