8th World Congress on Structural and Multidisciplinary Optimization

1. Abstract Over three decades, metamodeling has been widely applied to design optimization problems to build a surrogate model of computation-intensive engineering models. The Kriging method has gained significant interests for developing the surrogate model. However, traditional Kriging methods, including the ordinary Kriging and the universal Kriging, use fixed polynomials basis functions to generate the mean structure. In this paper, a new Dynamic Kriging (D-Kriging) method is proposed to fit the true model more accurately. In this D-Kriging method, the mean structure is automatically decided by applying sequential feature selection to the candidate basis functions based on a new accuracy criterion. In addition, a new sequential sampling technique based on the prediction interval of the surrogate model is proposed and integrated into the D-Kriging method. Numerical examples show much more accurate results from D-Kriging compared with traditional Kriging methods.