Multi-target Design Method Based On Expected Improvement

Meta materials attract extensive attention worldwide, however the design of there elements often consume large amount of time. Thousands of different elements’ parameters need to be chosen according to different target responses.Problems like finding designs corresponding to response targets are often faced in the industry. In many cases the response function of the design does not have an analytical form or even unknown, and getting the response of a design(taking an experiment) consumes large amount of time or money. Surrogate models like Kriging are often used to predict responses of designs we do not take any experiments, in order to avoid expensive experiments. Methods of design of experiment corresponding to single objective and multiple objectives are widely discussed in literature, however sometime problems unlike single and multiple objective problems, searching for different designs corresponding to different response values from a single function, like our problem,are faced in the industry. In this article, Kriging model is used to a real database from a problem of metamaterial design, and a new design of experiment method for multiple target values problem is discussed. A test using Branin function shows that the new method is much more efficient and stable than using single objective method separately to every target value.