Expedited Simulation-Driven Optimization Using Adaptively Adjusted Design Specifications

The focus of this chapter is the adaptively adjusted design specifications (AADS) technique, which, in some sense, is qualitatively different from the SBO methods discussed so far in this book. The key component of the majority of the SBO techniques is a surrogate model, constructed so as to represent well (in a given sense that might be problem dependent) the high-fidelity model being the subject of the optimization process. In the case of physics-based models, the surrogate is created through a suitable correction of an underlying low-fidelity model. A number of correction techniques have been discussed in detail in the preceding chapters of this book. The perspective offered by AADS is different: instead of correcting the low-fidelity model it modifies the design requirements for the problem at hand. The modifications are implemented to account for the discrepancies between the low- and the high-fidelity models. Given that the models are sufficiently well correlated, optimization of the (intact) low-fidelity model leads to design improvement of the high-fidelity one. There is no need for any model correction, which is an important advantage of AADS, therefore making it simple to implement. On the other hand, AADS can only be used for certain types of problems, including those for which the design specifications can be expressed in a minimax form. In this chapter, we formulate the AADS methodology and illustrate its use for solving various design problems related to computational electromagnetics, such as design of microwave filters, antennas, and high-frequency transition structures.