Study of some strategies for global optimization using Gaussian process models with application to aerodynamic design
Abstract
Aerodynamic shape optimization using CFD and global optimizers like PSO is a computationally expensive process. To make the method more efficient, metamodels must be incorporated into the optimization algorithm. Metamodels can be either local or global in their approximation ability. We use local models based on kriging in a pre-evaluation strategy to screen a small number of promising designs, which are then evaluated on the exact model. We also construct and use global metamodels as a complete replacement of the exact model, with various enrichment methods based on merit functions. The two methods are applied to inviscid, transonic drag reduction of a wing under lift and volume constraints, and their performance is compared. In all the tests, the use of metamodels reduces the computational expense. Global metamodels were found to give the maximum reduction in computational expense.
Origin | Files produced by the author(s) |
---|
Loading...