Frontiers in Energy Research (Jun 2020)

Development of Multi-GPU–Based Smoothed Particle Hydrodynamics Code for Nuclear Thermal Hydraulics and Safety: Potential and Challenges

  • So-Hyun Park,
  • Young Beom Jo,
  • Yelyn Ahn,
  • Hae Yoon Choi,
  • Tae Soo Choi,
  • Su-San Park,
  • Hee Sang Yoo,
  • Jin Woo Kim,
  • Eung Soo Kim

DOI
https://doi.org/10.3389/fenrg.2020.00086
Journal volume & issue
Vol. 8

Abstract

Read online

Advanced modeling and analysis are always essential for the development of safe and reliable nuclear systems. Traditionally, the numerical analysis codes used for nuclear thermal hydraulics and safety are mostly based on mesh-based (or grid-based) methods, which are very mature for well-defined and fixed domains, both mathematically and numerically. In support of their robustness and efficiency, they have been well-fit into many nuclear applications for the last several decades. However, the recent nuclear safety issues encountered in natural disasters and severe accidents are associated with much more complex physical/chemical phenomena, and they are frequently accompanied by highly non-linear deformations. Sometimes, this means that the conventional methods encounter many difficult technical challenges. In this sense, the recent advancement in the Lagrangian-based CFD method shows great potential as a good alternative. This paper summarizes recent activities in the development of the SOPHIA code using Smoothed Particle Hydrodynamics (SPH), a well-known Lagrangian numerical method. This code incorporates the basic conservation equations (mass, momentum, and energy) and various physical models, including heat transfer, turbulence, multi-phase flow, surface tension, diffusion, etc. Additionally, the code newly formulates density and continuity equations in terms of a normalized density in order to handle multi-phase, multi-component, and multi-resolution problems. The code is parallelized using multiple graphical process units (GPUs) through multi-threading and multi-streaming in order to reduce the high computational cost. In the course of the optimization of the algorithm, the computational performance is improved drastically, allowing large-scale simulations. For demonstration of its applicability, this study performs three benchmark simulations related to nuclear safety: (1) water jet breakup of FCI, (2) LMR core melt sloshing, and (3) bubble lift force. The simulation results are compared with the experimental data, both qualitatively and quantitatively, and they show good agreement. Besides its potential, some technical challenges of the method are also summarized for further improvement.

Keywords