Scholarlink

The Point Kernel method is widely regarded as one of the most effective techniques in radiation shielding design and analysis. To ensure its capabilities for rapid radiation field calculations, the complex photon scattering effects are simplified and integrated into buildup factors, the accuracy of which directly influences the reliability of shielding calculations. The most referenced dataset for buildup factors originates from the ANS 6.4.3-1991 report; however, this dataset has been withdrawn due to lack of updates and maintenance. Additionally, the data were derived from numerical simulations that did not account for all photon reactions, and the dataset is limited to a narrow range of physical quantities. In this study, a newly developed buildup factors database is presented, generated using the Monte Carlo code RMC, which incorporates all relevant photon reaction effects. The new database adopts a multi-energy-group format, thereby broadening its applicability and enhancing its physical interpretability. Furthermore, deep learning-based calculation models, as a replacement of the original GP method, are proposed for unified and user-friendly applications. Various data processing techniques and two different loss criterions are evaluated to optimize performance. This work aims to extend the utility of the traditional Point Kernel method and expand its applicability in nuclear engineering practice.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Preliminary Deep Learning Framework for Predicting Multi-group Photon Fluence Buildup Factors in Radiation Shielding

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Junyi Chen,
  • Chenghao Cao,
  • Ruihan Li,
  • Ben Qi,
  • Jingang Liang

摘要

The Point Kernel method is widely regarded as one of the most effective techniques in radiation shielding design and analysis. To ensure its capabilities for rapid radiation field calculations, the complex photon scattering effects are simplified and integrated into buildup factors, the accuracy of which directly influences the reliability of shielding calculations. The most referenced dataset for buildup factors originates from the ANS 6.4.3-1991 report; however, this dataset has been withdrawn due to lack of updates and maintenance. Additionally, the data were derived from numerical simulations that did not account for all photon reactions, and the dataset is limited to a narrow range of physical quantities. In this study, a newly developed buildup factors database is presented, generated using the Monte Carlo code RMC, which incorporates all relevant photon reaction effects. The new database adopts a multi-energy-group format, thereby broadening its applicability and enhancing its physical interpretability. Furthermore, deep learning-based calculation models, as a replacement of the original GP method, are proposed for unified and user-friendly applications. Various data processing techniques and two different loss criterions are evaluated to optimize performance. This work aims to extend the utility of the traditional Point Kernel method and expand its applicability in nuclear engineering practice.