An adjoint-response surface method is developed to give global representation of cost function in a parametrized design space for turbomachinery blades. Radial basis function (RBF) based and quadratic polynomial (QP) based response surface models are constructed using both the values of cost function and its adjoint gradients with respect to geometry control parameters. The method is tested on a quasi-three dimensional NACA0012 blade row, then applied to the transonic Rotor 67. In preliminary design optimization stage, when the number of undetermined control parameters is large, the QP based model can provide a global image of the cost function in high dimensional design space with a small amount of sample points. In two-parameter fine optimization stage, high resolution can be achieved with the RBF based models. This gradient-enhanced response surface method is useful in guiding designers to discover the global optimum which may be missed by local gradient methods in a complicated design space. It may also be used as substitute of CFD flow solver in time consuming iterative design and optimization.
Mixing Global and Local Competition in Genetic Optimization based Design Space Exploration of Analog Circuits