Molecular modeling, simulation and machine learning studies of polymer nanocomposite structures and morphologies
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Polymer nanocomposites (PNCs) comprised of nanoscale fillers embedded in a polymer matrix exhibit enhanced mechanical properties, electrical conductivity, improved optical, and/or thermal properties as compared to the neat matrix polymer. These enhanced macroscopic properties of PNCs are strongly dependent on the structure and assembled morphology of these PNC constituents arising from their chemistry and interactions and material processing conditions. Molecular modeling and simulations serve as valuable computational tools that can complement experiments by exploring effects of various PNC design parameters and shedding light on the molecular mechanism driving the assembly and phase behavior in PNCs. Machine learning (ML) approaches also serve as valuable computational tools aiding experiments by providing objective and automated interpretation of structural characterization (e.g., scattering, microscopy) of PNCs. The overarching goal of this doctoral thesis is the application and development of coarse-grained modeling and simulations and machine learning workflows to accelerate the exploration of design-structure-morphology relationship of PNCs, specifically containing nanorod / nanowire as nanofillers. ☐ In the first part of this dissertation, I describe the development of coarse-grained models for simulating nanorod morphologies in PNCs. In this part, I present results from these simulations showing how the two types of nanorod surface functionalization - homogeneous functionalization and heterogeneous functionalization- affect nanorod morphologies in PNCs. I also explored the effect of nanorods’ physical roughness on the resulting nanorod morphologies in PNCs. In both studies, the work describes how the propensity of nanorod dispersion, nanorod aggregation and/or percolation are affected by the surface functionalization and physical roughness. The results are rationalized through arguments of the competing enthalpic and entropic driving forces in the nanocomposites. ☐ In the second part of this dissertation, I shift focus towards data-driven approaches for understanding experimental structural characterization, specifically electron microscopy and small angle X-ray scattering (SAXS), results. In the first study in this part of the dissertation, I developed a semi-supervised machine learning (ML) workflow for data-efficient multi-task analysis of microscopy images containing different morphologies of protein nanowires. Multi-task analysis includes classification of the nanowire wire morphology and segmentation of the location of the nanowires from the microscopy images. In the second study in this data-driven methodology development, I created a ML workflow called PairVAE (Paired Variational AutoEncoder) to generate one type of structural characterization (e.g. microscopy image) from another type of structural characterization (e.g., SAXS profile) as input. I trained the PairVAE model on open-access SAXS and scanning electron microscopy (SEM) characterization data of block copolymer thin film material assembled through template-directed self-assembly. After proper training, I demonstrated that PairVAE can generate an SEM image from the SAXS data and vice versa resembling the corresponding morphological features. ☐ Overall, in this thesis, I describe research work of predictive molecular modeling and simulations of nanorod morphologies in PNCs and development of ML workflow for accelerating structural characterization of polymer materials. This thesis work provides guidelines and outlook for applied ML towards automation and acceleration of soft material characterization data interpretation and reconstruction.
Description
Keywords
Molecular modeling, Simulations, Polymer nanocomposites, Machine learning, Nanofillers