These days robust design has become more and more important to raise reliability in the products under many kinds of impression, including human errors, modeling errors and so on. In the past, robust design only tries to obtain the robust design variables that have less sensitivity with quasi-optimum solutions. Recently, many researchers try to design not only of design variable's value but also to its deviations. Which means that robust design tries to have wider range of design variables and still keeps robustness in the design. For the simplicity of calculation, most of them assume independence of design variables, thus, the ranges are determined by their upper and lower bound. However, in the real design almost every design variables are not independent with the influence of constraints. In this study, we will newly propose the method that can consider relationships of design variables, something like correlation of design variables. We will show the effectiveness of the method by some numerical examples.