Documenti full-text disponibili:
Abstract
This thesis investigates the modeling, prediction, and optimization of wireless power transfer (WPT) systems operating in electromagnetically complex environments, addressing both Industrial Internet of Things (IIoT) applications and biomedical wireless powering platforms. While full-wave electromagnetic (EM) simulations provide accurate field characterization, their computational cost becomes prohibitive in large-scale and obstacle-rich scenarios. To address this challenge, physics-based EM modeling is combined with artificial intelligence techniques to enable efficient and scalable electromagnetic field prediction. A computationally efficient Integral Equation–based framework is first employed to generate power-density distributions in confined industrial environments. Active learning strategies are introduced to reduce the number of required full-wave simulations while preserving representative coverage of the configuration space. Building upon this dataset, artificial neural networks are developed to predict both optimal sensor locations and complete power-density maps. Convolutional neural networks are shown to effectively capture the spatial structure of EM fields, achieving superior performance in near-field prediction tasks. To further improve scalability, a physics-informed residual learning framework is proposed, combining superposition-based field approximations with neural correction of interaction effects neglected by the superposition process. This approach reduces the dependence on exhaustive full-wave simulations while maintaining physical consistency. Incremental learning experiments further demonstrate the generalization capability of the proposed methodology, showing that models trained on simpler environments can be successfully adapted to more complex obstacle configurations. In parallel, the thesis applies electromagnetic optimization principles to the design of a battery-free wireless optogenetic stimulation platform, enabling reliable wireless power delivery in heterogeneous biological environments. Overall, the work demonstrates how electromagnetic theory and data-driven intelligence can be combined to enable efficient prediction and optimization of wireless power systems in both industrial and biomedical applications.
Abstract
This thesis investigates the modeling, prediction, and optimization of wireless power transfer (WPT) systems operating in electromagnetically complex environments, addressing both Industrial Internet of Things (IIoT) applications and biomedical wireless powering platforms. While full-wave electromagnetic (EM) simulations provide accurate field characterization, their computational cost becomes prohibitive in large-scale and obstacle-rich scenarios. To address this challenge, physics-based EM modeling is combined with artificial intelligence techniques to enable efficient and scalable electromagnetic field prediction. A computationally efficient Integral Equation–based framework is first employed to generate power-density distributions in confined industrial environments. Active learning strategies are introduced to reduce the number of required full-wave simulations while preserving representative coverage of the configuration space. Building upon this dataset, artificial neural networks are developed to predict both optimal sensor locations and complete power-density maps. Convolutional neural networks are shown to effectively capture the spatial structure of EM fields, achieving superior performance in near-field prediction tasks. To further improve scalability, a physics-informed residual learning framework is proposed, combining superposition-based field approximations with neural correction of interaction effects neglected by the superposition process. This approach reduces the dependence on exhaustive full-wave simulations while maintaining physical consistency. Incremental learning experiments further demonstrate the generalization capability of the proposed methodology, showing that models trained on simpler environments can be successfully adapted to more complex obstacle configurations. In parallel, the thesis applies electromagnetic optimization principles to the design of a battery-free wireless optogenetic stimulation platform, enabling reliable wireless power delivery in heterogeneous biological environments. Overall, the work demonstrates how electromagnetic theory and data-driven intelligence can be combined to enable efficient prediction and optimization of wireless power systems in both industrial and biomedical applications.
Tipologia del documento
Tesi di dottorato
Autore
Augello, Elisa
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
Artificial Neural Network, Wireless Power Transfer, Electromagnetically Harsh Environments, Internet of Things
DOI
10.48676/unibo/amsdottorato/13291
Data di discussione
6 Luglio 2026
URI
Altri metadati
Tipologia del documento
Tesi di dottorato
Autore
Augello, Elisa
Supervisore
Co-supervisore
Dottorato di ricerca
Ciclo
38
Coordinatore
Settore disciplinare
Settore concorsuale
Parole chiave
Artificial Neural Network, Wireless Power Transfer, Electromagnetically Harsh Environments, Internet of Things
DOI
10.48676/unibo/amsdottorato/13291
Data di discussione
6 Luglio 2026
URI
Statistica sui download
Gestione del documento: