| タイトル | Propagation of Computational Uncertainty Using the Modern Design of Experiments |
| 本文（外部サイト） | http://hdl.handle.net/2060/20080002264 |
| 著者（英） | DeLoach, Richard |
| 著者所属（英） | NASA Langley Research Center |
| 発行日 | 2007-12-03 |
| 言語 | eng |
| 内容記述 | This paper describes the use of formally designed experiments to aid in the error analysis of a computational experiment. A method is described by which the underlying code is approximated with relatively low-order polynomial graduating functions represented by truncated Taylor series approximations to the true underlying response function. A resource-minimal approach is outlined by which such graduating functions can be estimated from a minimum number of case runs of the underlying computational code. Certain practical considerations are discussed, including ways and means of coping with high-order response functions. The distributional properties of prediction residuals are presented and discussed. A practical method is presented for quantifying that component of the prediction uncertainty of a computational code that can be attributed to imperfect knowledge of independent variable levels. This method is illustrated with a recent assessment of uncertainty in computational estimates of Space Shuttle thermal and structural reentry loads attributable to ice and foam debris impact on ascent. |
| NASA分類 | Statistics and Probability |
| 権利 | No Copyright |
