PNNL

Abstract

Uncertainty is a key metric in computational modeling that must be evaluated for results to have wide ranging applicability. A well characterized uncertainty range is ideal with clear error bars on results that can be presented to stakeholders. In the field of spent fuel cask modeling, this ideal has been historically difficult to achieve in practice because of the computationally intensive nature of the models used and the difficulty assigning reasonable uncertainties to quantities in as-built systems. The work in this report has been conducted to evaluate the overall state of uncertainty and sensitivity in spent fuel cask models and develop methodologies for evaluating these uncertainties. These methodologies must be practical for engineering applications. They should not require excessive computational resources or calendar time to achieve results. In engineering, the model must be on a scale such that it can be changed and adapted throughout a project as new information is discovered and project goals evolve. This report covers three major modeling task areas that provide an overview of the types of sensitivity and uncertainty present in a spent fuel storage and transportation system. Section 3 discusses sensitivity and uncertainty analysis in the effective thermal conductivity model for the fuel region and applies these results to a single assembly model. Section 4 shows sensitivity analysis of a full cask model in the TN-32B and section 5 demonstrates the overall uncertainty workflow using Coolant Boiling in Rod Arrays – Spent Fuel Storage and STAR-CCM+ developed from the sensitivity work in the preceding sections. Key outcomes and demonstrations from this work include: • The results of the effective thermal conductivity uncertainty quantification show that this quantity, while important to modeling, has minimal effect on overall uncertainty and can be estimated reliably with different codes. • Key sensitivities were identified in the TN-32B High Burnup Demonstration Research Project Cask; these will inform uncertainty analysis and future transient modeling. • Latin hypercube sampling uncertainty quantification was demonstrated as a practical method for uncertainty quantification even with computationally intensive full cask models. • Full cask model uncertainty results showed good agreement with data and demonstrated the importance of input parameter distribution selection. • Methodology for a streamlined workflow utilizing multiple analysis tools was developed and can be applied to any future spent fuel cask modeling or other relevant systems that can be computationally modeled. The work in this report was undertaken to study sensitivities and uncertainties in spent fuel cask modeling and then develop methodologies for uncertainty analysis that can be practically applied to future work. These goals were successfully met and there is now a demonstrated workflow in place that the Department of Energy can use when analyzing thermal models and can that be applied to other relevant computational modeling areas.