2021/08/05 by Jaber Rezaei Mianroodi, Mianroodi, Jaber Rezaei, Shahed Rezaei +7 · 3 citations
Materials Science · Biochemistry, Genetics and Molecular Biology · Engineering · #Machine Learning in Materials Science #Advanced Electron Microscopy Techniques and Applications #Composite Material Mechanics
paper · pdf · doi:10.48550/arxiv.2108.02837
The elastic properties of materials derive from their electronic and atomic\nnature. However, simulating bulk materials fully at these scales is not\nfeasible, so that typically homogenized continuum descriptions are used\ninstead. A seamless and lossless transition of the constitutive description of\nthe elastic response of materials between these two scales has been so far\nelusive. Here we show how this problem can be overcome by using Artificial\nIntelligence (AI). A Convolutional Neural Network (CNN) model is trained, by\ntaking the structure image of a nanoporous material as input and the\ncorresponding elasticity tensor, calculated from Molecular Statics (MS), as\noutput. Trained with the atomistic data, the CNN model captures the size- and\npore-dependency of the material's elastic properties which, on the physics\nside, can stem from surfaces and non-local effects. Such effects are often\nignored in upscaling from atomistic to classical continuum theory. To\ndemonstrate the accuracy and the efficiency of the trained CNN model, a Finite\nElement Method (FEM) based result of an elastically deformed nanoporous beam\nequipped with the CNN as constitutive law is compared with that by a full\natomistic simulation. The good agreement between the atomistic simulations and\nthe FEM-AI combination for a system with size and surface effects establishes a\nnew lossless scale bridging approach to such problems. The trained CNN model\ndeviates from the atomistic result by 9.6 % for porosity scenarios of up to\n90 % but it is about 230 times faster than the MS calculation and does not\nrequire to change simulation methods between different scales. The efficiency\nof the CNN evaluation together with the preservation of important atomistic\neffects makes the trained model an effective atomistically-informed\nconstitutive model for macroscopic simulations of nanoporous materials and\nsolving of inverse problems.\n