Design Optimization of a Loudspeaker Utilizing Sampling-Based Sensitivity Information of a Hyperspherical Local Window

This paper proposes an efficient optimization method that can deal with electromagnetic design problems with high-dimensional design variables. To achieve this, the universal Kriging method is combined with a new-type local window, called the hypersphere, and the truncated Gaussian sampling. At the center of a nominal design point, accurate surrogate models for performance functions of interest are generated in a relatively small hyperspherical local window. The first-order design sensitivity values are extracted from Jacobian matrices of the locally approximated models. Such optimization scheme considerably reduces sampling points and iterative designs, and also facilitates deriving the design sensitivity with high precision. The proposed method is tested with two design models having more than 10 design variables. Its efficiency and accuracy is thoroughly investigated by comparing it with existing methods.