DETERMINATION OF CRITICAL EXPERIMENT CORRELATIONS VIA THE MONTE CARLO SAMPLING TECHNIQUE

Critical benchmark experiments are the foundation of validation of the computational codes used in criticality safety analyses because they provide a basis for comparison between the calculated results and the physical world. These experiments are often performed in series varying a limited number of parameters to isolate the effect of the independent parameter. The use of common materials, geometries, machines, procedures, detectors, or other shared features can create correlations among the resulting experiments. Most validation techniques used in criticality safety practice do not treat these correlations explicitly, and the effect of this is unclear as the correlations themselves are not well known. Generalized linear least squares methods used for advanced validation or in data adjustment studies also rely on correlation coefficients to constrain the adjustments allowed in critical experiment results. The purpose of this dissertation is to develop a methodology for the calculation of critical experiment correlations using a Monte Carlo sampling technique. The use of this technique allows for the determination of the uncertainty in each individual experiment, and identical perturbations applied to shared parameters provide estimates of the covariance between the experiments. The correlation coefficient is then calculated by dividing the covariance between any pair of experiments by the product of the individual experiment standard deviations. This technique is applied to high-enriched uranium solution experiments and low-enriched uranium pin lattice experiments to determine correlation coefficients for these types of systems. The important parameters governing the correlation coefficients are determined, and the results are compared with correlation coefficients in the literature determined using other methods at other institutions. The general method for the determination of the correlation coefficients is presented along with other conclusions and recommendations for further study in this area.