Machine Learning Approach for Transforming Scattering Parameters to Complex Permittivity

TL;DR

This paper presents a neural network model that accurately converts scattering parameters into complex permittivity values for granular catalysts, streamlining dielectric property estimation across a broad frequency range.

Contribution

The study introduces a supervised convolutional neural network approach that directly predicts complex dielectric properties from scattering parameters, eliminating iterative solution challenges.

Findings

High accuracy in predicting dielectric constants from scattering data

Effective validation with experimental measurements

Model covers a wide frequency and dielectric constant range

Abstract

This study investigates the application of an artificial neural network to predict the complex dielectric properties of granular catalysts commonly used in microwave reaction chemistry. The study utilizes finite element electromagnetic simulations and two-dimensional convolutional neural networks to solve for a large solution space of varying dielectrics. This convolutional neural network was trained using a supervised learning approach and a common backpropagation. The frequency range of interest was between 0.1 to 13.5 GHz with the real part of the dielectric constants ranging from 1 to 100 and the imaginary part ranging from 0.0 to 0.2. The network was double validated using experimental data collected from a coaxial airline. The model was demonstrated to convert either experimental or computational derived scattering parameter to complex permittivities. Moreover, the model eliminates the need for iterative solutions that often have difficulty with the piecewise continuous nature of frequency dependent scattering parameters.