Abstract
Metamodelling based search, space exploration, and region reduction/elimination methods are effective optimization schemes for computation intensive global design optimization problems. In this work a new metamodelling, space exploration and region reduction search algorithm is introduced. This algorithm, namely Space Exploration and Unimodal Region Elimination (SEUMRE), divides the design space into key unimodal regions using design experiment data; identifies the regions that most likely contain the global minimum; fits Kriging models with additional design experiments using Latin Hypercube designs over these regions; identifies their local minima, and then the global optimum. By identifying promising unimodal regions of the objective and reducing searching space, the method can find the global optimum effectively and efficiently, particularly suited for optimization problems that require extensive computation through engineering analyses and simulations. Comparisons with existing space exploration and region elimination/reduction methods using benchmark test problems have been carried out to demonstrate the advantages of the new method. More robust and problem independent metamodelling improvements are under study.