Identifying interphase properties in polymer nanocomposites using adaptive optimization

To predict the properties of nanocomposites, computational models have demonstrated that the interphase behavior can be expressed using the matrix properties with a modified (shifted) frequency. The amount of the shift necessary for a given sample data set can be determined by achieving the best fit between the predicted curve from a computation model and the experimental data through trial-and-error. However, with the complexity of experimental data and expensive computational costs, a manual process to solve this inverse problem is impractical to handle many experimental data sets. This difficulty hinders investigation of the underlying principles behind nanocomposite interphase. In this work, we present an adaptive optimization approach that accelerates the search for interphase properties in polymer nanocomposite data sets by solving the inverse problem using global optimization. The objective is to minimize the difference between the predicted bulk property of a nanocomposite with that from the experiment data. A Gaussian Process (GP) model is built as a surrogate of the objective function with quantification of prediction uncertainty. An adaptive sampling strategy is applied to effectively navigate the complex search space by iteratively selecting the next sampling point based on an expected improvement function. The surrogate model and the optimal solution evolve until the desired objective is achieved. The approach is tested on both the simulations of dielectric and viscoelastic properties in nanocomposites. Our work provides insight into identifying the interphase properties for polymer nanocomposites using adaptive optimization and demonstrates the potential of data-driven approach for achieving a deeper understanding of the interphase properties and its origins.