Prediction of H2 Adsorption Energy on Graphene Oxide Flakes Using Neural Networks

This thesis aims to use two different Neural Networks to predict the adsorption enegy of hydrogen molecules on graphene oxide nanostructures. The first, called Graphenet, is a pretrained neural network which works with RGB images. The second one is a custom Convolutional Neural Network which overcomes the 3-channel costraint given by the image representation, modifying the input dataset creating 6,8 and 10 channel input dataset. For both methods discussed, most of the work involves modifying the input dataset to make it as broad and meaningful as possible in order to achieve the best prediction accuracy.