Using Information about Experimental Conditions to Predict Properties of Metamaterials

Abstract

In this work, a method of increasing the amount of data for training neural networks is proposed using the possibility of using information about the experimental conditions of measuring the properties of metamaterials. It is shown that the method is flexible and effective. The results of predicting the transmission coefficient of the metamaterial for different angles of incidence of radiation and type of polarization are presented. Using the architecture presented in the work, a high rate of learning and generation of new data was obtained with an error that does not exceed 12% for experiments in one frequency range and does not exceed 31% if all experiments are used for training. The architecture of the neural network and the method by which it is possible to easily change the number and types of experimental conditions are presented.

Keywords

• metamaterials
• 3D convolutional neural network
• experimental conditions