Scholarlink

Generating large-scale datasets on supercomputers is a critical component of modern research, enabling sophisticated analysis and mining applications. In the context of SUBOFF model studies for advanced submarine development, we emphasize the significance of supercomputer-generated big data and its subsequent application. Utilizing the Sunway TaihuLight supercomputer and the SWLBM software, we employ the lattice Boltzmann method (LBM) to generate a comprehensive dataset of flows over a SUBOFF model. This dataset is then used to train a physics-informed neural network (PINN) model, which aims to reconstruct flow fields from sparse velocity measurements. The PINN is designed to perform super-resolution velocity reconstruction and concurrent pressure field inference. Our results show that the reconstructed flow fields, including pressure, are in good agreement with the full-resolution LBM references, highlighting the promise of this approach for complex flow motion reconstruction. This research demonstrates the potential of supercomputer-generated big data in enhancing simulations and experiments in the study of SUBOFF and beyond.

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

Large-Scale Data Generation Using SWLBM on the Sunway TaihuLight Supercomputer and Subsequent Data Mining with Physics-Informed Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xuesen Chu,
  • Wei Guo,
  • Tianqi Wu,
  • Shengze Cai,
  • Guangwen Yang

摘要

Generating large-scale datasets on supercomputers is a critical component of modern research, enabling sophisticated analysis and mining applications. In the context of SUBOFF model studies for advanced submarine development, we emphasize the significance of supercomputer-generated big data and its subsequent application. Utilizing the Sunway TaihuLight supercomputer and the SWLBM software, we employ the lattice Boltzmann method (LBM) to generate a comprehensive dataset of flows over a SUBOFF model. This dataset is then used to train a physics-informed neural network (PINN) model, which aims to reconstruct flow fields from sparse velocity measurements. The PINN is designed to perform super-resolution velocity reconstruction and concurrent pressure field inference. Our results show that the reconstructed flow fields, including pressure, are in good agreement with the full-resolution LBM references, highlighting the promise of this approach for complex flow motion reconstruction. This research demonstrates the potential of supercomputer-generated big data in enhancing simulations and experiments in the study of SUBOFF and beyond.