Propagation of Computational Uncertainty Using the Modern Design of Experiments

摘要: This paper describes the use of formally designed experiments to aid in error analysis a computational experiment. A method is described by which underlying code approximated with relatively low-order polynomial graduating functions represented truncated Taylor series approximations true response function. resource-minimal approach outlined such can be estimated from minimum number case runs code. Certain practical considerations are discussed, including ways and means coping high-order functions. The distributional properties prediction residuals presented discussed. for quantifying that component uncertainty attributed imperfect knowledge independent variable levels. illustrated recent assessment estimates Space Shuttle thermal structural reentry loads attributable ice foam debris impact on ascent.