MPACT Efficiency and Robustness Enhancements for Full-Core BWR Modeling

University of Michigan

https://dx.doi.org/10.7302/4143

NURAM-2021-002-00

MPACTBoiling Water ReactorBWRneutron transportnuclear reactor physics

Because of the Boiling Water Reactor (BWR)’s unique features such as the cruciform control blade, coolant void, two-phase flow, and heterogeneous geometry, the performance and robustness of MPACT for BWR applications is degraded compared to its performance for Pressurized Water Reactor (PWR) applications. In order to improve the efficiency and robustness forBWRsimulations, several enhancements were considered and implemented in the work performed here, including: • adoption of the linear-source MOC; • improved iterative methods; • memory reduction by using mixed single and double precision; • optimization of problem initialization. The Linear Source Approximation (LSA) and the optimized meshing were initially developed for PWR applications, and the LSA has not been used routinely for whole core calculations. In this work, issues with the linear source with respect to its robustness and optimal meshing for BWRs have been addressed. For the new iteration scheme, the multilevel in energy CMFD solver was combined with a sophisticated feedback-based partial convergence technique. The method was first shown to be very effective in reducing the number of outer iterations and MGCMFD iterations for multiphysics PWR applications, and was then adapted to BWR applications in this work. Although some considerations for robustness remain to be resolved. The mixed precision technique combined the use of different numerical precisions in the MPACT computational algorithm in order to reduce memory usage. Several variables with a large memory footprint that are not directly related to convergence checks (:eff, fission source) are now stored as single-precision reals and converted back to double precision only in the calculation. All three enhancements improve the efficiency of MPACT for BWR simulations. The robustness of LSA and the new iteration scheme is improved in order to realize these efficiency gains. Finally, the problem initialization process has been optimized to speed up the geometry and meshing set-up at the beginning of a problem. The new iterative methods have been shown to speed up the coupled MPACT simulation of the Peach Bottom 2 (PB2) cycle 1 problem by a factor of 2. The LSA and mixed precision reduce the total memory by 17% for the PB2 problem with a minimum runtime impact. The optimization of geometry and meshing setup results in a speedup of 30-40% in the problem initialization, depending on the number of unique assemblies and control cells. The remainder of the runtime speed up is attributed to the MEDPC algorithm.

Technical Report