Neutron analysis of spent fuel storage installation using parallel computing and advance discrete ordinates and Monte Carlo techniques
Daniel Shedlock,Alireza Haghighat
DOI: https://doi.org/10.1093/rpd/nci226
Abstract:In the United States, the Nuclear Waste Policy Act of 1982 mandated centralised storage of spent nuclear fuel by 1988. However, the Yucca Mountain project is currently scheduled to start accepting spent nuclear fuel in 2010. Since many nuclear power plants were only designed for -10 y of spent fuel pool storage, > 35 plants have been forced into alternate means of spent fuel storage. In order to continue operation and make room in spent fuel pools, nuclear generators are turning towards independent spent fuel storage installations (ISFSIs). Typical vertical concrete ISFSIs are -6.1 m high and 3.3 m in diameter. The inherently large system, and the presence of thick concrete shields result in difficulties for both Monte Carlo (MC) and discrete ordinates (SN) calculations. MC calculations require significant variance reduction and multiple runs to obtain a detailed dose distribution. SN models need a large number of spatial meshes to accurately model the geometry and high quadrature orders to reduce ray effects, therefore, requiring significant amounts of computer memory and time. The use of various differencing schemes is needed to account for radial heterogeneity in material cross sections and densities. Two P3, S12, discrete ordinate, PENTRAN (parallel environment neutral-particle TRANsport) models were analysed and different MC models compared. A multigroup MCNP model was developed for direct comparison to the SN models. The biased A3MCNP (automated adjoint accelerated MCNP) and unbiased (MCNP) continuous energy MC models were developed to assess the adequacy of the CASK multigroup (22 neutron, 18 gamma) cross sections. The PENTRAN SN results are in close agreement (5%) with the multigroup MC results; however, they differ by -20-30% from the continuous-energy MC predictions. This large difference can be attributed to the expected difference between multigroup and continuous energy cross sections, and the fact that the CASK library is based on the old ENDF/B-II library. Both MC and SN calculations were run in parallel on a BEOWULF PC-cluster (eight processors). Timing results indicate that the SN calculation yielded a detailed dose distribution at over 318,426 points in -164 h. Unbiased continuous energy MC required 214 h to calculate dose rates with a 1% relative error in only 18 regions on the surface of the cask. The biased A3MCNP calculations yields dose rates with -0.8% relative error in only 2.5 h on one processor. This study demonstrates that a parallel code, such as the 3-D parallel SN transport code, PENTRAN can solve a complex large problem, such as the storage cask, accurately and efficiently. Moreover, this calculation was performed on a relatively inexpensive PC-cluster. Possible inadequacies of the CASK cross section library still need to be evaluated.