Dr. Tomasz Kozlowski
Nuclear, Plasma, and Radiological Engineering
University of Illinois at Urbana-Champaign
The statistical uncertainty methodologies, such as the ones used by DAKOTA and RAVEN code, are well developed and commonly used in research and industry. The initial step of such uncertainty methodology is selection of the uncertain input parameters and their probability distribution functions (PDFs). The uncertainty assigned to each input parameter is typically done through expert judgement. Since most of the physical models use correlations with best-fit coefficients, it is vital that uncertainty quantification on these physical model coefficients is carried out to represent uncertainty of the simulation results. Bayesian-based inverse methods are proposed to estimate input uncertainty and reduce reliance on expert judgement. The seminar will demonstrate inverse uncertainty quantification of input models identified and ranked as highly sensitive to the figure of merit (e.g., peak clad temperature) and forward uncertainty quantification for prediction of safety-significant event (e.g., post-dryout).
Materials Science & Engineering Dept.