A Benchmark Study of Computer Codes for System Analysis of the Nuclear Fuel Cycle

Abstract

As use of nuclear energy is expected to expand in different parts of the world, several codes that describe the nuclear fuel cycle system are currently under development, featuring a range of capabilities and different levels of flexibility and automation. Such codes model the addition or retirement of reactors, the demand for fresh fuel, and the need for spent fuel storage and recycling facilities as the production of nuclear energy varies with time. The codes enable analysis of various scenarios for the evolution of the nuclear energy system, and the timing of deployment of new facilities. Outputs may also include fuel material mass flows, economic analysis and metrics related to spent fuel or waste assessment. The study reported here is the first attempt for benchmarking the MIT code CAFCA against three independently developed fuel cycle simulation codes. It is also among the first publicly available benchmark exercises. Some reviews of the existing codes were previously reported, but focused mostly on their theoretical capabilities. Benchmarking studies, generally involving two or three codes, have been done over the last few years, but most remain unpublished. The codes included in this study are: CAFCA (developed at MIT), COSI (developed at CEA, France), DANESS (developed at ANL) and VISION (developed by DOE laboratories for the AFCI program). The purpose of this benchmark study is to evaluate the degree of convergence of the current versions of these codes and to compare their basic methodologies. This benchmark is not a comprehensive analysis of all the codes’ capabilities but constitutes a first step towards a more complete benchmark study. In order to compare all 4 codes, only the common capabilities were considered and assessed. Those capabilities are essentially those of CAFCA, as it is the simplest code. Consequently, some of the advanced capabilities of the other codes were disabled, and their complete features were not reflected in this benchmark study. In addition, economic evaluation of the fuel cycles was not considered, even though it is a capability common to the four codes. Furthermore, the initial runs showed that a degree of freedom should be removed to ease the comparison. For that reason reprocessing capacity profiles were provided by CAFCA and used as input by the other codes. Following a description of the codes, the report presents the four scenarios selected as the benchmark cases, including initial conditions. The time period of the simulation covers the 21st century. Those scenarios differ from each other in either the nuclear energy supply growth rate (0%, 1.5% or 3%) or the type of advanced technology introduced in the midterm. The options for advanced reactors were: the “self-sustaining” fast reactor (with a fissile conversion ratio of one), fast burners of transuranics (TRU), or a combination of plutonium recycling (as mixed oxide) in the thermal light water reactors (LWRs) and fast burners. The scenarios, and hence the results, are for benchmarking purposes only and should not be considered realistic for policy studies or forecasts about the future of nuclear power. The set of constraints specified is minimal and only intended to provide a common framework for the simulations. The results are presented and commented on for each case. The first case, characterized by a very constraining zero energy growth rate, shows an excellent agreement among the codes, with identical ratios of fast reactors/thermal reactors over time. This excellent agreement was the iv result of the particular efforts made in order to get very close results (several iterations were performed to allow for adjustments). This case eventually shows how the models can produce very close results if sufficiently tuned to adhere to the same basic assumptions. This case also allowed us to identify a number of minor apparent discrepancies and explain them. In particular, it made obvious differences in results between COSI, which was tuned to track fuel batches (“discrete-flow code”) and CAFCA, DANESS and VISION, which deal with annual mass flows (”continuous-flow codes). The treatment of discrete batches of fuel by COSI, instead of timeaveraged quantities in the other codes, result in somewhat oscillatory flows and inventories of materials. Another factor leading to discrepancies among the codes is the time assumed to exist between the separation of transuranics and the manufacturing of fast reactor fuels. CAFCA speeds up the fuel manufacturing, to avoid the presence of separated transuranics in large quantities, while the other codes do not (as Pu-containing fuel has a very limited shelf-life at the initial fissile content due to Pu 241 decay with a half life of about 14.4 years). Unlike the first cases, there was in the three other cases no attempt, beyond the common set of assumptions, to iterate to get the results of the other cases to converge. Therefore, these three cases are more a reflection of how the codes actually operate and show the level of variation in results that should be considered normal. These three cases were of great interest for comparing the different strategies for fast reactor deployment and their dependence for fuel on available TRU from the operation of light water reactors. Overall, the benchmark shows that, although the codes exhibit reassuring consistency, both internally and among themselves, differences still exist. The fact that reprocessing capacity profiles were externally provided may have disturbed some of the codes designed to internally calculate this variable. Although limitations inherent in the codes exist, the differences in the results generally do not reveal major flaws, but rather reflect differing assumptions and constraints embedded in the methods and approximations of the calculations. This benchmark reveals (or reminds us) that there is no single profile for a fuel cycle scenario but several profiles that depend on industrial practices with regards to manufacturing, storage and reprocessing of nuclear fuels. These practices may aim at differing priorities of reducing stocks of stored spent fuel, avoiding the presence of separated TRU to reduce proliferation risks, ensuring sufficient supply of fresh fuel for advanced reactors and spent fuel for reprocessing plants, and minimizing some costs. Such choices can either be intrinsic to the code (through built-in assumptions) or through user choices (for example, the level of conservatism in the algorithm ensuring fuel supply for fast reactors is a user input). Moreover, the parameters left to the user’s discretion are generally not the same from one code to another, or are expressed in different terms. Finally, complete consistency between the codes is difficult to obtain. Two major conclusions can be drawn from this benchmark. First, the overall results show good consistency and similar trends. Hence, utilization of various codes is likely to lead to similar conclusions. Second, one must not expect the various fuel cycle system simulation codes to provide identical outputs. Therefore, users must keep in mind that, although the results are internally consistent and meet each code’s set of requirements, they do not project unique scenarios for meeting such requirements.