Full Core modeling techniques for research reactors with irregular geometries using Serpent and PARCS applied to the CROCUS reactor

This paper summarizes the results of modeling methodologies developed for the zero-power (100 W) teaching and research reactor CROCUS located in the Laboratory for Reactor Physics and Systems Behavior (LRS) at the Swiss Federal Institute of Technology in Lausanne (EPFL). The study gives evidence that the Monte Carlo code Serpent can be used effectively as a lattice physics tool for small reactors. CROCUS' core has an irregular geometry with two fuel zones of different lattice pitches. This and the reactor's small size necessitate the use of nonstandard cross-section homogenization techniques when modeling the full core with a 3D nodal diffusion code (e.g. PARCS). The primary goal of this work is the development of these techniques for steady-state neutronics and future transient neutronics analyses of not only CROCUS, but research reactors in general. In addition, the modeling methods can provide useful insight for analyzing small modular reactor concepts based on light water technology. Static computational models of CROCUS with the codes Serpent and MCNP5 are presented and methodologies are analyzed for using Serpent and SerpentXS to prepare macroscopic homogenized group cross-sections for a pin-by-pin model of CROCUS with PARCS. The most accurate homogenization scheme lead to a difference in terms of k eff of 385 pcm between the Serpent and PARCS model, while the MCNP5 and Serpent models differed in terms of k eff by 13 pcm (within the statistical error of each simulation). Comparisons of the axial power profiles between the Serpent model as a reference and a set of PARCS models using different homogenization techniques showed a consistent root-mean-square deviation of $8%, indicating that the differences are not due to the homogenization technique but rather arise from the definition of the diffusion coefficients produced by Serpent. A comparison of the radial power profiles between the best PARCS model and full-core Serpent model showed largest relative differences in terms of power prediction at the core periphery, which is believed to be the product of the geometry simplifications made, the diffusion coefficients produced by Serpent, and the two-group energy structure used. The worth of a single control rod reproduced in PARCS showed a difference of À33 pcm from its 169 pcm worth simulated in Serpent.