Current Awardee

Discovery of Advanced Nanodielectrics through AI-accelerated Multiscale Simulation from First Principles

Yumeng Li
Yumeng Li

College: Grainger College of Engineering
Award year: 2019-2020
NCSA collaborators: Erman Guleryuz

The research objective of this proposal is to create a new generation artificial intelligence-enabled multiscale simulation framework and its enabling techniques, which would directly root in first principles theory, for a comprehensive understanding of the cross-scale multiphysics phenomena in dielectric polymer nanocomposites. In addition to features like easy processing and lightweight, polymer nanocomposites demonstrate a great potential in realizing highly enhanced combined properties to meet the needs of advanced dielectrics in applications from energy storage to power delivery. However, the current lack of understanding in fundamental mechanisms leading to the property enhancement necessitates modeling of complex phenomena (ranging from nanoscale to macroscale) using a high-performance multiscale simulation framework. The new multiscale simulation framework employs artificial intelligence (AI) for an effective integration of first principle calculations, physics-based atomistic simulations and data-driven predictive analytics, thereby concurrently leveraging high accuracy of first principles calculations and high efficiency in AI enabled predictive data analytics. Built upon the PI's research experience on both multiscale simulation and polymer nanocomposites, this proposal will focus on four research thrusts to address grand challenges in developing the new framework: 1) developing machine learning potentials for the interface based on high-throughput first principles calculations 2) characterizing nanoscale local interfacial electro-mechanical-thermal properties using AI-accelerated atomistic simulations, 3) predicting macroscale electro-mechanical-thermal properties considering the interface effects, and 4) model validation of the proposed multiscale simulation framework.