Accelerated Feature-Based Local Optimization with Variable-Fidelity EM Simulations

This chapter discusses an approach to accelerated feature-based local optimization with the use of variable-fidelity EM simulation models. The considered acceleration mechanism constitutes one of the available methods of expediting local FBO procedures. Other options include the employment of adjoint sensitivities as well as sparse sensitivity updates, which will be considered in Chap. 8. Utilization of variable-fidelity EM simulations is directly aimed at lowering the computational cost of the optimization process. This is achieved by delegating some of the computational effort, normally executed using high-fidelity EM analysis, into a cheaper low-fidelity model. In the majority of variable-fidelity optimization frameworks, only two levels of fidelities are exploited: coarse/fine and low-/high-fidelity models. The low-fidelity model may be based on coarse-discretization EM analysis (typically in antenna design), but also equivalent networks (more often to be utilized in microwave design). The examples of this chapter utilize the former type. In a majority of frameworks, the discretization density of the coarse model is adjusted using engineering experience and visual inspection of the family of EM-simulated responses of the structure under design, evaluated at different levels. Appropriate adjustment of the model fidelity is not a trivial task: setting discretization too low may lead to a failure of the optimization process. Yet, increasing the discretization overly may compromise the cost efficacy of the process. The FBO frameworks of this chapter utilize dedicated procedures for an automated determination of the discretization level of a low-fidelity model to ensure that the trade-off between reliability and computational cost is maintained. Specifically, in the considered algorithms, low-fidelity EM simulations are employed for the evaluation of the system gradients, thereby allowing us to achieve a considerable speedup of the optimization process. The discussed frameworks are demonstrated using several examples of antennas and microwave structures. Comprehensive benchmarking and application case studies, along with experimental verification, are also provided.