Atomistic simulation and machine learning predictions of mechanical response in nanotube-polymer composites considering filler morphology and aggregation

Hamid Ghasemi,Hessam Yazdani

DOI: https://doi.org/10.1016/j.commatsci.2024.113399

IF: 3.572

2024-10-19

Computational Materials Science

Abstract:Pursuing innovative materials through integrating machine learning (ML) with materials informatics hinges critically upon establishing accurate processing-structure–property-performance relationships and consistently applying them in training datasets. Pivotal to unraveling these relationships is an accurate representation of the microstructure in computational models. In this study, we use transmission electron microscopy (TEM) micrographs of carbon nanotubes (CNTs) within a polymer matrix to construct representative polymer-nanotube composite (PNC) models. We then simulate the models using the coarse-grained molecular dynamics (CG-MD) technique to elucidate the influence of filler morphology and aggregation on the mechanical properties of PNCs. Besides CNTs, we consider cyanoethyl nanotubes (C 3 NNT) as a representative of the carbon nitride family, which has remained largely unexplored as a PNC filler for load-bearing purposes. We employ the CG-MD results to train ML models—neural network (NN), support vector regression (SVR), and Gaussian process regression (GPR)—to predict the strain–stress responses of PNCs. Results indicate the profound influence of the filler morphology and aggregation on the elastic and shear stiffness of PNC composites. A high degree of transverse isotropy is observed in the mechanical behavior of composites with perfectly oriented fillers, with Poisson's ratios surpassing conventional upper bounds observed in isotropic materials. For a given morphology, C 3 NNT composites exhibit higher stiffness in longitudinal and transverse directions than CNT composites. The ML models demonstrate accuracy in predicting the strain–stress response of the composites, with the GPR model showing the highest accuracy, followed by the NN and SVM models. This accuracy makes the ML models readily integrable into a multiscale modeling framework, significantly enhancing the efficiency of transferring information across scales.

materials science, multidisciplinary