Computational analysis of polymer nanocomposites across length scales

Abstract

Polymer nanocomposites (PNC) exhibit excellent mechanical properties making them a promising material for several engineering applications. The enhancement in properties is primarily attributed to the toughening and stiffening mechanisms provided by nanofillers. Computational techniques have proven to be a powerful tool for studying the properties and behavior of materials. Since the dimensions of the components of PNC ranges from the nano to macroscale, a computational study on PNCs should account for mechanics across multiple length scales. The primary objective of this thesis is to investigate the fundamental mechanisms that govern the mechanical properties of PNCs. A better understanding of these mechanisms can facilitate the efficient design of PNCs for desired applications. To this end, this thesis is structured into two parts. First, we employ molecular dynamics (MD) simulations to study the effect of functionalization and agglomeration on the mechanical properties of PNCs. Second, we investigate the relevance of multiscale simulations and higher-order continuum theories as alternatives to computationally expensive MD simulations.

Functionalization and agglomeration are two key factors which influence the mechanical properties — especially fracture properties — of PNCs. Through a suite of MD simulations, we studied the influence of varying degrees of functionalization on the elastic and fracture properties. Interestingly, it was found that there exists an optimal degree of functionalization corresponding to maximum enhancement in elastic property, tensile strength, ductility, and fracture toughness. The underlying mechanics behind this optimality is identified through careful studies on crack propagation mechanisms, including crack arresting and the formation of new crack surfaces. Moreover, the improvement in interfacial interaction induced by functionalization was explained from an atomistic perspective. A sequence of similar studies on ag- glomerated PNCs revealed that even functionalization does not improve fracture toughness. The agglomeration results in crack to completely propagate through the polymer matrix. Thus the CNTs do not participate in resisting the crack. Furthermore, the shear strain distribution along the CNT surface revealed that the CNTs and the polymer surrounding them act as a monolithic unit, resulting in a reduction in fracture toughness and ductility. The agglomerated CNTs are also found to contribute towards strain accumulation and failure of PNC at a lower global strain level. These atomistic studies provide valuable insights into the overall mechanical behavior of PNCs. However, the computational cost of MD simulations hinders its application to large systems.

In the second part of the thesis, we study possible alternatives to expensive MD simulations. An MD-informed hierarchical multiscale method was adopted to estimate the effective elastic properties. Through this method, the elastic properties of PNC and hierarchical PNC were determined at a low volume percentage of CNT reinforcement. The results were in good agreement with the published experimental values. Attaining such a low volume fraction using MD simulations is impossible due to the computational cost. Following the idea of reducing the computational cost, the relevance of higher-order continuum theories in predicting the mechanical response of nanotubes was investigated. However, these theories have unknown length scale parameters which need to be computed. For this purpose, we used MD and ML in tandem. Initially, MD simulations were used to generate a dataset consisting of different variables and the corresponding length scale parameter values. Then ML-based techniques were used to make further predictions. The method proved to be effective in eliminating further MD simulations within the limit of the dataset generated. However, the alternatives to MD simulation presented in this thesis are confined to estimating the elastic properties.

In general, this thesis provides a comprehensive computational study at different length scales of a PNC. Several atomic-level mechanisms instigated due to functionalization and agglomeration are presented. A few alternatives to elude the computational cost of MD simulations are also discussed.