Title:Deep learning framework for carbon nanotubes: mechanical properties and modeling strategies

Abstract:Tensile tests at room temperature are performed using molecular dynamics on all configurations of single-walled carbon nanotubes up to 4 nm in diameter. Distributions of the Young's modulus, Poisson's ratio, ultimate tensile strength and fracture strain are determined and reported. The results show that the chirality of the nanotube has the greatest influence on the properties. An artificial neural network is developed for the dataset obtained by molecular dynamics and used to predict the mechanical properties. It is clearly shown that Deep Learning provides accurate predictions, with the further advantage that thermal fluctuations are smoothed out. In addition, a through analysis of the effect of dataset size on prediction quality is performed, providing modeling strategies for further researchers.