Evaluating Experimental Designs for Fitting Response Surfaces of Deterministic Models.
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING
Pagination or Media Count:
This thesis evaluates several bias minimizing and variance minimizing experimental designs in terms of their effectiveness and efficiency in constructing response equations for a deterministic, nuclear exchange problem. The criteria which is used to evaluate these designs includes 1 number of required design points 2 number of terms in the response equation 3 accuracy of fit of the response equation 4 orthogonality of the design and 5 rotatability of the design. In addition, the response surface equations are evaluated in terms of their predictive power and their explanatory power. The predictive power addresses the equations ability to adequately estimate the true surface and to accurately predict a future response for a given set of inputs. The explanatory power addresses the equations ability to present a response equation which is simple to interpret so that the true surface can be easily evaluated and the results can be easily explained. Keywords Response surface methodology Experimental design Deterministic models.
- Statistics and Probability