민감도 정보를 이용한 설계 방법 및 소프트웨어의 개발

Abstract

Sensitivity information has been used for linearization of nonlinear functions in optimization. Basically, sensitivity is a derivative of a function with respect to a design variable. Design sensitivity is repeatedly calculated in optimization. Since sensitivity calculation is extremely expensive, there are studies to directly use the sensitivity in the design process. When a small design change is required, an engineer makes design changes by considering the sensitivity information. Generally, the current process is performed one-by-one for design variables. Methods to exploit the sensitivity information are developed. When a designer wants to change multiple variables with some relationship, the directional derivative can be utilized. In this case, the first derivative can be calculated. Only small design changes can be made from the first derivatives. Orthogonal arrays can be used for moderate changes of multiple variables. Analysis of Variance is carried out to find out the regional influence of variables. A flow is developed for efficient use of the methods. A software system with the flow has been developed. The system can be easily interfaced with existing commercial systems through a file wrapping technique. The sensitivity information is calculated by finite difference method. Various examples are solved to evaluate the proposed algorithm and the software system.