![[Plasma and Fusion Research]](/PFR/pfr_header.gif)
Plasma and Fusion Research
Volume 21, 2403017 (2026)
Regular Articles
- 1)
- Graduate School of Science and Engineering, Yamagata University, Yonezawa 992-8510, Japan
- 2)
- Department of Research, National Institute for Fusion Science, National Institutes of Natural Sciences, Toki 509-5292, Japan
- 3)
- Graduate School of Engineering, Nagoya University, Nagoya 464-8603, Japan
- 4)
- Simulation Engineering Division, Toyota Technical Development Corporation, Toyota 470-0334, Japan
- 5)
- Graduate School of Science and Technology, Shinshu University, Nagano 380-8553, Japan
- 6)
- Department of Applied Physics and Physico-Informatics, Keio University, Yokohama 223-8522, Japan
- 7)
- National Institutes for Quantum Science and Technology, Rokkasho 039-3212, Japan
- 8)
- National Institutes for Quantum Science and Technology, Naka 311-0193, Japan
- 9)
- National Institutes of Technology, Nagaoka College, Nagaoka 940-0817, Japan
Abstract
Understanding hydrogen trapping in tungsten is crucial for accurate modeling of plasma-wall interactions in fusion devices. In this study, we developed a deep learning model to predict hydrogen trapping sites, which serve as essential input parameters for kinetic Monte Carlo (kMC) simulations. The model employs a U-Net–based convolutional neural network that directly maps three-dimensional potential energy distributions to trapping site positions. Training data were generated from atomistic calculations using the embedded atom method (EAM) potential, and ground-truth trapping sites were systematically identified by force relaxation. The trained model achieved an F1 score of approximately 0.76, with most predicted sites coinciding with the true minima within ± 1 voxels (1 voxel = 0.1 Å per side). Visual comparisons confirmed the ability of the model to capture both global and local features of the potential energy landscape. In terms of efficiency, the proposed approach reduced prediction time by more than three orders of magnitude compared with conventional force-relaxation searches, enabling predictions in less than one second on a GPU. These results demonstrate that deep learning provides an accurate and computationally efficient method for identifying trapping sites. Future extensions include incorporating more complex defect structures and integrating the model into molecular dynamics-kMC hybrid frameworks for large-scale plasma-wall interaction simulations.
Keywords
deep learning, U-Net, dice loss, atomistic simulations, hydrogen trapping sites, tungsten, plasma-wall interactions, MD, kMC, fusion materials
Full Text
References
- [1] G. Janeschitz et al., J. Nucl. Mater. 220-222, 73 (1995).
- [2] G.F. Matthews, J. Nucl. Mater. 220-222, 104 (1995).
- [3] B. Lipshulz et al., J. Nucl. Mater. 220-222, 967 (1995).
- [4] C.S. Pitcher and P.C. Stangeby, Plasma Phys. Control. Fusion 39, 779 (1997).
- [5] S.I. Krasheninnikov et al., Phys. Lett. A 214, 285 (1996).
- [6] K. Ohsawa et al., J. Nucl. Mater. 458, 187 (2015).
- [7] S. Saito et al., Nucl. Fusion 64, 126067 (2024).
- [8] S. Saito et al., Nucl. Mater. Energy 43, 101942 (2025).
- [9] S. Saito et al., Contrib. Plasma Phys. e201900152 (2020).
- [10] S. Saito et al., Plasma Fusion Res. 15, 2403073 (2020).
- [11] H. Nakamura et al., Jpn. J. Appl. Phys. 61, SA1005 (2022).
- [12] S. Saito et al., Jpn. J. Appl. Phys. 60, SAAB08 (2021).
- [13] K. Sawada et al., Contrib. Plasma Phys. e201900153 (2020).
- [14] S. Saito et al., Jpn. J. Appl. Phys. 63, 09SP03 (2024).
- [15] O. Ronneberger et al., MICCAI 2015, LNCS 9351, 234 (2015).
- [16] Ö. Çiçek et al., MICCAI 2016, LNCS 9901, 424 (2016).
- [17] F. Milletari et al., Proceedings of 2016 Fourth International Conference on 3D Vision (3DV), (IEEE, 2016), 565–571.
- [18] L.F. Wang et al., J. Phys. Condens. Matter. 29, 435401 (2017).