Innovative defect cluster analysis algorithm and tool powered by machine learning techniques

Reactor pressure vessel materials are critical to the safety and longevity of nuclear reactors, with irradiation damage posing significant challenges to their performance. Understanding defect clusters is essential to elucidating material behavior under irradiation. This paper makes four key contributions: (1) it employs the union-find algorithm to efficiently segment defect clusters in molecular dynamics data; (2) it integrates machine learning with material defect cluster characteristics, proposing a novel algorithm for defect cluster identification and classification, enabling detailed cluster visualization; (3) it introduces a defect data filtering method based on normal distribution, improving defect cluster classification accuracy; and (4) it develops software for analyzing defect clusters in nuclear reactor materials based on the proposed algorithms. The software is used to investigate the relationships between primary knock-on atom (PKA) energy, PKA direction, and defect cluster types and quantities, yielding results consistent with experimental observations. The proposed algorithm and software simplify the complex process of material defect analysis, providing an efficient and intuitive tool for advancing material mechanism research.