Robust variable-fidelity optimization of microwave filters using co-Kriging and trust regions

This work introduces a variable-fidelity optimization methodology for simulation-driven design optimization of filters. Our approach is based on electromagnetic (EM) simulations of different accuracy controlled by the mesh density. A Kriging interpolation model (the surrogate) is created using sampled low-fidelity EM data and optimized to approximately locate the optimum of the high-fidelity EM model of the filter. This initial surrogate is subsequently improved by blending in the high-fidelity data accumulated during the optimization process using the co-Kriging technique. The algorithm convergence is ensured by embedding it into the trust region framework. The operation and performance of our method is demonstrated using three filter design cases.