불용 초기점을 갖는 변수 선별 정보를 이용한 다중 응답 시스템의 대체모델 기반 최적설계 프레임워크

Abstract

In designing multiple response system (MRS), surrogate model-based design optimization (SMBDO) is often performed because it takes a lot of time and cost for simulation analyses. However, as the number of design variables increases, prediction accuracy of the surrogate models decreases, and the number of function calls to construct the surrogate models increases. Variable screening techniques that select design variables with significant influence on the responses are being implemented to solve the issues. The surrogate model with high prediction accuracy can be constructed by using the variable screening information of the MRS having infeasible initial points, however, infeasible solution can be derived through optimization. Furthermore, if the surrogate models are reconstructed by adding the significant design variables to derive feasible solution, a serious loss of time and cost is caused by additional sampling and simulation analyses as the design domain of the surrogate models are changed. Therefore, this study proposes a SMBDO framework that can be used without waste of sample points used for variable screening and surrogate modeling when infeasible solution is derived.