크리깅 메타모델의 외삽 예측성 평가

Abstract

Carrying out optimum design, in general, designer needs to set up ranges of design variables at the beginning. At this point, after we fulfil optimization by using metamodel and design of experiment, we need to modify the ranges of design variables or additional sample point if initial optimum point is located around boundary that designer set arbitrarily. In this study, through extrapolation of metamodel applied to initial design, we quantitatively evaluate accuracy of predicted responses outside of the boundary and propose ranges of extrapolation keeping accuracy without additional sample points. We select krigingmodel as the metamodel which gives Bayesian information and evaluates expanding performance after we examine the nonlinearity according to R-square of response surface model(RSM) through mathematical examples.