Shahsavaripour Atefe, Badiei Mohammad Hossein, Kalhor Ahmad, Yousefi Leila
Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, 1417614411, Iran.
School of Engineering and Informatics, University of Sussex, Falmer, UK.
Sci Rep. 2025 Mar 19;15(1):9426. doi: 10.1038/s41598-025-94116-9.
Metamaterial-based electromagnetic absorbers, despite being thin and lightweight, typically suffer from narrow-band frequency bandwidth and sensitivity to polarization and incident angle due to their resonant nature. Previous methods to increase bandwidth have shown improvements but have not fully succeeded in developing wide-band, thin metamaterial-based absorbers suitable for mass production. In this study, we introduce a novel approach that leverages artificial intelligence to design a thin, wideband metamaterial-based absorber covering the entire frequency range of 8-12 GHz. The proposed method utilizes a Generative Adversarial Network (GAN), given the need for precise structural details and computational efficiency, which globally outperform variational autoencoders (VAEs) and diffusion models, for parameter estimation and a Multi-Layer Perceptron (MLP) network as a simulator to predict the electromagnetic response of the designed absorber and provide feedback to the generative network. Numerical full-wave electromagnetic simulations serve as the training data and ground truth for both the GAN and MLP networks. This training enables the generative network to produce structures with high absorption, while the MLP predicts the corresponding absorbance value for each structure. This approach allows for the rapid design of various real-world structures, quick calculation of their absorption values using the MLP network, and selection of the most optimal structures for fabrication. The performance of the designed metamaterial-based absorber is verified both numerically and experimentally. Results show an absorption rate above 90% for all frequencies in the range of 8-12 GHz. The structure also operates effectively for both TE and TM polarizations and for all incident angles between 0-45 degrees. Additionally, the designed structure can be easily fabricated using printed circuit board (PCB) technology, making it practical and suitable for mass production.
基于超材料的电磁吸收器尽管薄且轻，但由于其共振特性，通常存在窄带频率带宽以及对极化和入射角敏感的问题。以往增加带宽的方法虽有改进，但在开发适用于大规模生产的宽带、薄型超材料吸收器方面尚未完全成功。在本研究中，我们引入了一种新颖的方法，利用人工智能设计一种薄型宽带超材料吸收器，其覆盖8 - 12 GHz的整个频率范围。鉴于需要精确的结构细节和计算效率，所提出的方法利用生成对抗网络（GAN），其在全局上优于变分自编码器（VAE）和扩散模型，用于参数估计，并使用多层感知器（MLP）网络作为模拟器来预测所设计吸收器的电磁响应，并向生成网络提供反馈。数值全波电磁模拟用作GAN和MLP网络的训练数据及基准事实。这种训练使生成网络能够生成具有高吸收率的结构，而MLP则预测每个结构的相应吸光度值。这种方法允许快速设计各种实际结构，使用MLP网络快速计算其吸收值，并选择最优化的结构进行制造。所设计的基于超材料的吸收器的性能通过数值和实验得到验证。结果表明，在8 - 12 GHz范围内的所有频率下，吸收率均高于90%。该结构对于TE和TM极化以及0 - 45度之间的所有入射角均有效运行。此外，所设计的结构可以使用印刷电路板（PCB）技术轻松制造，使其实用且适合大规模生产。
