Deep Learning Assisted Large Scale Metamaterial Simulation
Date
2024
Authors
Advisors
Journal Title
Journal ISSN
Volume Title
Repository Usage Stats
views
downloads
Abstract
Simulating large scale electromagnetic (EM) problems has always been a challenging task and has no standard solutions which works effectively for a wide range of cases. These types of problems are frequently encountered in real-world applications and hold significant value as large-scale artificial electromagnetic devices offer increased flexibility in achieving exceptional performance targets. To cope with this challenge, this dissertation centers on proposing methods for effectively simulating large-scale electromagnetic problems by incorporating the concept of the Green's function and the emerging powerful tool of deep learning. On one hand, the effectiveness of using known analytic Green's functions for large scale problems is first presented. The unavailability of Green's functions for most problems hinders its broader application. To achieve this goal, a generalized Green's function is introduced and deep neural networks (DNNs) are employed to efficiently represent these generalized Green's functions, which serve the same purpose in solving large-scale EM problems as traditional Green's functions. However, DNNs introduce another challenge that can be equally difficult as solving the problem with numerical methods due to the demanding process of data preparation and training. To this end, transfer learning techniques are introduced to generalize the DNN based method for broader range of similar problems. The proposed techniques of generalized Green's functions and deep neural networks (DNNs) both make significant contributions to solving large-scale problems. Finally, applications using an experimentally measured Green's functions in large-scale microwave imaging systems are demonstrated, suggesting potential future experimental and practical applications of DNN-based generalized Green's functions beyond electromagnetic problem simulation.
In summary, I proposed and implemented different simulation methods with Green's functions or deep learning which all converge towards the ultimate goal of effectively simulating large-scale EM problems. These methods are demonstrated to assist with solving complex EM problems in terms of computation time, data requirements or the capability of generalization to a whole class of problems. The DNN-based generalized Green's function indicates a viable path to effectively solving specific large scale EM problems. I also experimentally conducted two different types of microwave imaging involving unconventional metamaterials with measured Green's function. These experiments illustrate that a thorough understanding of the Green's function of complex systems can lead to practical applications, including imaging.
Type
Department
Description
Provenance
Subjects
Citation
Permalink
Citation
Peng, Rixi (2024). Deep Learning Assisted Large Scale Metamaterial Simulation. Dissertation, Duke University. Retrieved from https://hdl.handle.net/10161/30795.
Collections
Except where otherwise noted, student scholarship that was shared on DukeSpace after 2009 is made available to the public under a Creative Commons Attribution / Non-commercial / No derivatives (CC-BY-NC-ND) license. All rights in student work shared on DukeSpace before 2009 remain with the author and/or their designee, whose permission may be required for reuse.