Abstract
Optimum design of expensive calculation model has been required. Response surface methodology is developed to solve the expensive calculation model by approximation models. Although the approximation models constructed by the response surface methodology are usually used to get the optimal designs, it is thought that the design accuracy is dependent on the type of activate functions and the design region of interest. In this paper, techniques to search all the local optimal designs within the feasible design region, and techniques for finding more accurate approximation using the proposed Most Probable Optimal Design (MPOD) method is introduced. The application example of the Hybrid III dummy model shows that the MPOD technique is effective in saving the computational cost.
