Improved modeling of the university of wisconsin nuclear reactor by automatic generation of computational models

Computational models for the University of Wisconsin Nuclear Reactor (UWNR) exist primarily to support the design of experiments, including an impending measurement of transient reactor physics behavior. Prior modeling efforts faced inherent challenges on long term maintainability and usability. Therefore, an improved model has been built with the following features: (1) documentation with reliable source of data, (2) build consistency by virtue of automation, (3) human readability and transparency of implemented changes on the model, (4) configuration management by version control, and (5) modularity for alternate representations. The benefits of these improvements are examined in terms of model accuracy, flexibility of modeling choices, and predictive analysis of the reactor conditions. MCNP6 demonstrated a computational bias of 604 ± 48 pcm, and Serpent showed an additional bias of 28 ± 23 pcm relative to the MCNP6 results. A study of the impact of varying modeling choices during burnup simulation found that the fuel discretization is important because of the higher fidelity representation of the temperature distribution, and therefore the variations in isotopic evolution. Estimates of the core life and power peaking factors were then generated using both MCNP6 and Serpent, using models automatically generated from the same source data.