Abstract

Surrogate models are becoming popular tools of choice in mitigating issues related to the excessive cost of electromagnetic (EM)-driven design of high-frequency structures. Among available techniques, approximation modeling is by far the most popular due to its versatility. In particular, the surrogates are exclusively based on the sampled simulation data with no need to involve engineering insight or problem-specific knowledge. Notwithstanding, a typically high nonlinearity of system outputs and the curse of dimensionality limit the applicability of conventional methods to relatively simple structures described by a few parameters within narrow ranges thereof. A recently reported nested kriging alleviates these difficulties from the perspective of an appropriate definition of the model domain. By focusing the modeling process on the region containing design that are optimized for the selected performance figures, it enables the construction of reliable surrogates over wide ranges of geometry/material parameters and operating conditions, both at a practically acceptable computational cost. The relative model domain thickness (i.e., its lateral-to-tangential size), determines the trade-off between the surrogate region of coverage and its predictive power, the former being essential for practical applications of the model, especially design optimization. This paper proposes a simple and computationally efficient procedure for automatic selection of the thickness parameter, which, in the original version of the method, was to be selected by the user. The importance of this aspect of the modeling process and the benefits of the proposed approach are demonstrated using a dual-band dipole antenna and miniaturized microstrip coupler.

Citations

Authors (2)