| About this Abstract |
| Meeting |
Materials Science & Technology 2020
|
| Symposium
|
Advances in Dielectric Materials and Electronic Devices
|
| Presentation Title |
Determining Complex Dielectric Properties from Coaxial Transmission Line Data Using a Machine Learning Approach |
| Author(s) |
Robert Tempke, Liam Thomas, Christina Wildfire, Dushyant Shekhawat, Terence Musho |
| On-Site Speaker (Planned) |
Robert Tempke |
| Abstract Scope |
This study investigated and developed an artificial neural network to predict the dielectric properties materials between 0.1-13.5 GHz. The approach utilized a two-dimensional convolutional neural network (CNN) in conjunction with a finite element electromagnetic model to generate a large solution space of different dielectric property combinations. This CNN was trained using a common back-propagation algorithm. The network is taught using supervised learning with a training, validation and test set. The dielectric material within the FE model was described using a complex description with the real part ranging from 1-100 and the imaginary part ranging from 0-0.2. Once convergence had been reached the network was double validated using experimental data collected in a coaxial airline. The same loss metrics were used to show that the network worked on experimental data and not just idealized computational data. |