Designing Multilayered Electromagnetic Absorbers with Machine Learning

This work presents a generative machine learning model that predicts upper graphene element configurations for a multilayer flexible electromagnetic absorber. The model uses the lower graphene layer as input and S-parameters as conditional input. It is increasingly difficult to fabricate multilayer absorbers using the conventional design process, which introduces performance and efficiency issues. With advances in machine learning algorithms and corresponding hardware, it is becoming easier to design multilayer absorbers using machine learning approaches. A conditional variational autoencoder—a type of generative machine learning model—was used to train our model. We integrated a residual network into our encoder portion to enable better feature extraction capabilities for our proposed model. We prepared the dataset using the High Frequency Structure Simulator, an electromagnetic simulation software program developed by Ansys. We reconstructed the upper and lower graphene structures from the dataset and predicted the upper graphene structure using the given S-parameters and the lower graphene structure of the absorber. Satisfactory agreement between the predicted and actual structures was observed.