Machine Learning–Based Prediction of Irradiation-Induced Swelling from Synergistic Effects of Helium and Hydrogen

Abstract

Understanding irradiation-induced degradation processes of nuclear structural materials is essential for creating methodologies and procedures for nuclear reactor safety. Due to the time- and resource-intensive property of both experiments and multiscale simulations of irradiation damage, the trial-and-error approach is completely inefficient. Recently, machine learning techniques have been employed to predict the properties of reduced activation ferritic martensitic (RAFM) steels, such as yield strength and elongation, as well as irradiation embrittlement in steel pressure vessels, with encouraging progress. In this work, void swelling is predicted using a machine learning method for the first time, taking into account the synergistic effects of displacement damage, helium, and hydrogen. Assisted by the analysis of feature engineering, seven machine learning models are trained and compared by multicriteria evaluation methods. Finally, the parameter-optimized gradient-boosting model is selected as the mapping function with the highest accuracy and universality to predict void swelling. In particular, the dependence of the void swelling and the injection amount of helium and hydrogen in the continuous parameter variation range is predicted beyond the existing experimental data. This work demonstrates the feasibility of machine learning to predict material irradiation damage by synergistic effects and has practical significance in nuclear material optimization and reactor safety.