Deep Learning for the Classification of Charge Density Plots of Lithium-Ion Cathode Materials

This paper proposes a novel approach to the classification of charge density plots of several types of lithium-ion cathode materials. Charge density plots visualize essential materials properties and are obtained as a result of the ab-initio atomic-scale materials supercomputer modeling. The paper presents a novel computational procedure for slicing the ab-initio simulated 3D data along each lattice vector and stepped on the grid points of computed material. Image datasets formed since the slicing procedure are classified with a deep learning convolution neural network, which is a pre-trained one, customized and fine-tuned after transfer learning. Computation experiments with the proposed deep learning classifier are performed and estimated.