For my year-long senior design project at Texas A&M, I conducted research for Dr. Braga-Neto to create a Physics-Informed Neural Network (PINN) in the field of Electromagnetics. A PINN is a neural network that uses a law of physics in the form of a nonlinear partial differential equation to train and solve the neural network. The use of a partial differential equation can replace the need for large data sets, and in some cases may not require any data, making these types of neural networks data efficient. The result is a neural network that can predict and simulate real world scenarios following the laws of physics. More information and examples about PINN's can be found here.

For my project, I used Maxwell's Equations to predict the far-field radiation patterns for a Yagi antenna since. The goal was to show that Maxwell's Equations could be used in a PINN to simulate a 3-D model of the electric and magnetic fields for an antenna. At the time of the project, there were no other examples of PINN's that used Maxwell's Equations for this type of application.

I used MATLAB's Antenna Toolbox to simulate a variety of Yagi antennas to create data sets for my model. To build the neural network, I used Tensorflow and created six general equations for the electric and magnetic fields in the x, y, and z dimensions to model Maxwell's Equations. The model was trained on the small set of simulated Yagi antennas. After training and predicting on the model, the data was modeled in MATLAB to display a spherical 3-D view of the radiation patterns with vector arrows to show direction and magnitude of each point for a single predicted antenna.

Since this project was for senior design, all research material is property of the school and is currently not available to the public. However, I can be reached by email to provide more information about the project.