Shayan Dodge
About Me
Shayan Dodge works at the intersection of computational electromagnetics and artificial intelligence. His research integrates high-fidelity numerical methods, including the finite element method (FEM), finite-difference time-domain (FDTD), and boundary element method (BEM), with physics-informed neural networks to develop scalable, physics-consistent surrogate models for electromagnetic systems.
His work aims to advance next-generation machine learning frameworks that enable real-time electromagnetic analysis and design optimization in complex multiphysics environments. His research is carried out under the supervision of Prof. Sami Barmada.
Key Contributions
Physics-Informed Neural Networks (PINNs) for Electromagnetic Problems
PINNs offer a promising alternative to classical electromagnetic (EM) solvers like FEM, FDTD, and BEM by learning solutions directly from governing equations, reducing the need for meshing and large-scale simulations. My research contributions include:
- Adaptive residual PINN (STAR-PINN) for time-domain magnetic diffusion [Paper][GitHub]
- Hybrid Boundary Element–PINN Method for Electromagnetic Analysis [Paper][GitHub]
Forecasting Lightning Effects in Electrical Systems (FELINES)
FELINES is a research project developing a preventive protection system for electrical infrastructures by sensing early lightning electromagnetic signals—especially Preliminary Breakdown Pulses (PBP)—to predict whether an upcoming Return Stroke (RS) will be dangerous, enabling timely disconnection of vulnerable equipment. Our contributions are documented in the following publications:
More details are available on the project website and GitHub.
AI-Based Optimization of Transcranial Magnetic Stimulation (TMS)
This work develops a data-driven framework to optimize transcranial magnetic stimulation (TMS) coil position and intensity using a deep learning model (variational autoencoder, VAE, combined with a convolutional neural network, CNN) trained on MRI-based realistic head models and FEM-based field distributions generated using a model order reduction (MOR) framework.
More details are available on the project website.
Recent News
- [4/2026]