Scholarlink

The flow region downstream of the spacer grid in a Pressurized Water Reactor (PWR) fuel rod bundle channel is characterized by intense shear effects and vortex structures, which significantly enhance lateral mixing and intensify heat transfer between the fuel rods and the coolant. However, due to the limitations of the Reynolds-averaged Navier-Stokes (RANS) equations, the computational accuracy of the flow field downstream of the spacer grid is insufficient. To achieve more accurate Computational Fluid Dynamics (CFD) simulation results, this paper proposes a neural network-based turbulence model optimization framework. By utilizing velocity data obtained from Particle Image Velocimetry (PIV) measurements of the fine-scale flow field downstream of the spacer gid, a Particle Swarm Optimization (PSO)-optimized turbulence closure model is developed. This optimized model is then embedded into RANS simulations to enhance the accuracy of the predicted velocity field in the downstream region of the fuel rod bundle channel. The results show that the proposed method significantly improves the RANS model’s performance in complex flow fields, yielding a velocity field that closely matches experimental measurements. The approach demonstrates the potential of coupling RANS and neural network models to improve computational performance in challenging flow scenarios.

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

Neural Network-Based Optimization for RANS Simulations of Flow Field in Fuel Bundle Channels with Spacer Grid

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Qishi Yin,
  • Wei Xu,
  • Denggao Chen,
  • Hui He,
  • Xiaojing Liu

摘要

The flow region downstream of the spacer grid in a Pressurized Water Reactor (PWR) fuel rod bundle channel is characterized by intense shear effects and vortex structures, which significantly enhance lateral mixing and intensify heat transfer between the fuel rods and the coolant. However, due to the limitations of the Reynolds-averaged Navier-Stokes (RANS) equations, the computational accuracy of the flow field downstream of the spacer grid is insufficient. To achieve more accurate Computational Fluid Dynamics (CFD) simulation results, this paper proposes a neural network-based turbulence model optimization framework. By utilizing velocity data obtained from Particle Image Velocimetry (PIV) measurements of the fine-scale flow field downstream of the spacer gid, a Particle Swarm Optimization (PSO)-optimized turbulence closure model is developed. This optimized model is then embedded into RANS simulations to enhance the accuracy of the predicted velocity field in the downstream region of the fuel rod bundle channel. The results show that the proposed method significantly improves the RANS model’s performance in complex flow fields, yielding a velocity field that closely matches experimental measurements. The approach demonstrates the potential of coupling RANS and neural network models to improve computational performance in challenging flow scenarios.