Polymer nanocomposites are highly tailorable materials that, with careful design, can achieve superior properties not available with existing materials. Most polymer nanocomposites are developed using an Edisonian (trial and error) process, severely limiting the capacity to optimize performance and increasing time to implementation. The solution is a data-driven design approach. As an example, this Designing Materials to Revolutionize and Engineer our Future (DMREF) project will design new material systems that simultaneously optimize for dielectric response and mechanical durability, a combination currently not achievable but necessary for high voltage electrical transmission and conversion. These new materials will have a significant economic impact on society because they will enable higher efficiency generation and transmission of electricity. More broadly, this new design approach will result in new nanostructured polymer material systems that will impact a wide range of industries such as energy, consumer electronics, and manufacturing. To ensure broad access to this work, the data, tools and models developed will be integrated and shared through an open data resource, NanoMine. The team will interact with the scientific community to create an integrated virtual organization of designers and researchers to test and improve the models. Educational components will reach undergraduate and graduate communities via interdisciplinary cluster programs at the two institutions, and provide undergraduate research opportunities and web based instructional modules and workshops.
The research is based on a central research hypothesis that using a data-driven approach, grounded in physics, allows integration of models that bridge length scales from angstroms to millimeters to predict dielectric and mechanical properties to enable the design and optimization of new materials. Data, algorithms and models will be integrated into the new and growing nanocomposite data resource NanoMine to address challenges in data-driven material design. This research will result in advancements in three areas. First, integrating a broad set of literature data and targeted experiments with multiscale methods will enable the development of interphase models to predict local polymer properties near interfaces considered critical for modeling polymer composites. Second, a hybrid approach utilizing machine-learning to bridge length scales between physics-based modeling domains will be used to create meaningful multiscale processing-structure-property relationship work flows. And, third, a Bayesian inference approach will utilize the knowledge contained in a dataset as a prior probability distribution and guide 'on-demand' computer simulations and physical experiments to accelerate the search of optimal material designs. Case studies will demonstrate the data-centric approach to accelerate the development of next-generation nanostructured polymers with predictable and optimized combinations of properties.
