On the use of LES-based turbulent thermal-stress models for rod bundle simulations

Nek5000 Enhancements for Faster Running Analysis

“…Nek5000's steady-state solvers, developed under the ECP CEED project, have been applied for solving steady thermal flows for a long rod bundle and pebble beds geometries. We demonstrate an improved performance for a rod bundle [4] and preliminary results for pebble bed simulations. We also describe the algorithmic approaches that are based on Schwarz preconditioned fast diagonalization methods [1,2] and Jacobian-Free Newton Krylov methods [3].…”

“…Our preconditioning strategy for a steady-state solver is based on Schwarz preconditioning with fast diagonalization, which enabled us to achieve 7× speedup for a long rod bundle case [1,2,4]. Adding this preconditioner to JFNK remains as a future work, which will improve performance further.…”

“…The idea reported here consists of using the information from the small LES calculation to determine the appropriate turbulence viscosity or turbulent thermal diffusivity that could be used to solve only for the temperature field in a pseudo-RANS approach. The detailed work has been published in IJHMT [4]; this report focuses on its subset, namely, the results by Nek5000's steady-state solver.…”

“…Considering the proposed methodology, once the average velocity field and the turbulent diffusivity are obtained by the pseudo-RANS approach, one may solve for the Reynolds-averaged Figure 1: Geometry for bare rod in triangular array [4]. L LES represents the length used for the pseudo-RANS approach, which is shorter than the full length L z .…”

“…Transient solver results are shown for the linear solver tolerances of 10 −2 (-), 10 −4 (-• -), and 10 −6 (---) with the timestep iterations and simulation time. The steady-state (SS) solver results are shown with the GMRES iterations and simulation time using E = spectral elements with N = 7 [4]. the simulations performed, considering the fastest time discretization scheme and larger tolerance (BDF3/EXT with tol = 10 −4 ), the SS solver is still about 7 times faster.…”