Sparse machine learning assisted deep computational insights on the mechanical properties of graphene with intrinsic defects and doping

2021 
Abstract Despite the tremendous capabilities of Molecular dynamics (MD) simulations, they suffer from the limitation of computationally intensive and time-consuming nature. This hinders the seamless discovery of nanomaterials with adequate computational insights. Over the last decade, graphene has received widespread attention from the scientific community due to its extra-ordinary multi-physical properties, primarily focusing on the fundamental physics and chemistry along with the notion of scalable synthesis. However, the recent advances in machine learning have opened new frontiers in the research of such exceptional two-dimensional materials where the boundaries of different multi-physical properties could be pushed further efficiently with the assistance of deep computational intelligence. Here we propose a coupled machine learning (ML) based approach to investigate the critical mechanical responses of graphene by capturing the underlying physics of the system through an (D-)optimally minimum number of MD simulations. We have investigated five different internal and external controlling input features like temperature, strain rate, nanostructural defects, doping, and chirality, which influence the critical mechanical properties of graphene such as the constitutive behaviour including fracture strength, failure strain and Young’s modulus. Even though the aspect of computational intensiveness in molecular dynamics simulations is addressed through the coupled ML based approach, in a data-driven comprehensive analysis, it is often difficult to get complete information about the concerning input parameters including their statistical distributions and occurrence bounds. Thus, it becomes difficult to carry out computational investigations based on powerful techniques like Monte Carlo simulation. To mitigate this lacuna, here we have exploited the ML-assisted approach for developing a level-based fuzzy framework at nano-scale under sparse input descriptions. In this article, we first demonstrate the computational efficiency achieved through the proposed ML based framework without compromising quality of the analysis, followed by data-intensive correlation analysis, sensitivity and uncertainty quantification considering various levels of the influencing system parameters, revealing detailed computational insights on mechanical properties of graphene including the coupled interactive influence of intrinsic defects and doping.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    81
    References
    1
    Citations
    NaN
    KQI
    []