Validation of Computer Models for Nuclear Material Shipping Packages

Computer models are abstractions of physical reality and are routinely used for solving practical engineering problems. These models are prepared using large complex computer codes that are widely used in the industry. Patran/Thermal is such a finite element computer code that is used for solving complex heat transfer problems in the industry. Finite element models of complex problems involve making assumptions and simplifications that depend upon the complexity of the problem and upon the judgment of the analysts. The assumptions involve mesh size, solution methods, convergence criteria, material properties, boundary conditions, etc. that could vary from analyst to analyst. All of these assumptions are, in fact, candidates for a purposeful and intended effort to systematically vary each in connection with the others to determine there relative importance or expected overall effect on the modeled outcome. These kinds of models derive from the methods of statistical science and are based on the principles of experimental designs. These, as all computer models, must be validated to make sure that the output from such an abstraction represents reality [1,2]. A new nuclear material packaging design, called 9977, which is undergoing a certification design review, is used to assess the capability of the Patran/Thermal computer model to simulate 9977 thermal response. The computer model for the 9977 package is validated by comparing its output with the test data collected from an actual thermal test performed on a full size 9977 package. Inferences are drawn by performing statistical analyses on the residuals (test data – model predictions).