A High Efficiency Cognitive-Driven Generative Design Method for Frequency Selective Surface

AI-assisted design is transforming the development of electromagnetic (EM) devices, moving away from the traditional trial-and-error approach to a more efficient and data-driven process. A cognitive-driven generative design method (CGDM) for optimizing discrete variable structure (DVS) devices is presented in this paper, which leverages easily interpretable response types as inputs. By addressing the inherent one-to-many mapping problem in inverse design, CGDM significantly reduces the reliance on large-scale training datasets. The integration of a confidence assessment network ensures the generation of multiple candidate solutions, providing optimal rankings based on confidence scores. Four design examples generated through CGDM demonstrate the method’s ability to generate high-quality solutions for various electromagnetic response types, including wideband, band-stop, low-pass, and band-pass FSS. These results were achieved with rapid generation and evaluation times, with confidence levels exceeding 92% for most categories and enabling robust selection of optimal designs from previously unseen patterns.