Georgia Institute of Technology Materials Science and Engineering

Committee Members:

Prof. Rampi Ramprasad, Advisor, MSE

Prof. David L. McDowell, ME/MSE

Prof. Juan-Pablo Correa-Baena, MSE

Prof. Seung Soon Jang, MSE

Prof. Roshan Joseph, ISyE

"Designing Polymers Resistant to Electric Field Extremes With Materials Modeling and Machine Learning"

Abstract:

Polymers have found applications as dielectrics in high energy density capacitors owing to their low cost, flexibility, attractive insulation properties, and ease of processability. However, their “energy density”, i.e., the maximum electrostatic energy that can be stored, is rather low in most commonly used polymer capacitor dielectrics; for instance, biaxially-oriented polypropylene (BOPP), the standard material used today in energy storage capacitors, displays an energy density of about 5 J/cc. The electrostatic energy density of a dielectric is directly controlled by its dielectric constant and the dielectric breakdown strength (Ebd), i.e., the maximum electric field the material can withstand. The goals of my work are to determine the factors that affect Ebd in polymers and to use this understanding to develop new polymers which can withstand large electric fields. The workflow used to achieve this can be divided into 4 steps: 

Identification of potential proxy properties (or “descriptors”) correlated to dielectric breakdown: Available phenomenological theories in conjunction with experimental data on dielectric breakdown for a number of benchmark polymers are utilized to reveal key descriptors. This step is essential because a search for polymers with high electric field resistance may be performed by devising screening criteria based on these proxy properties. Direct measurements or computation of the dielectric breakdown field for a large number of polymers is impractical at the present time, making the identification of the proxy properties critical.

Development of reliable computational methods to determine these proxy properties for a set of candidate polymers: Density functional theory (DFT) based methodologies are employed to gain a fundamental understanding of the relationships between chemistry and the proxy properties on the one hand (e.g., “design guidelines”), and to create a dataset which can be used to build machine learning-based prediction models of the proxy properties.

Development of machine learning (ML) based prediction models of the proxy properties: These “surrogate models” are trained on the proxy dataset and allow for the instantaneous predictions of the proxy properties for new cases. These models are used as a practical screening tool that can be aimed at a candidate set of polymers much larger than what DFT can handle. Finally, an AI-driven workflow was developed using the ML models for autonomous data creation using DFT.

Creation of a novel method to design synthetically accessible polymers:  A new approach to polymer design is introduced, starting from encoding easy-to-do, cheap experiments, and purchasable monomers. Further, the ML models of proxies are used to find high Ebd candidates from the synthesizable polymers generated using this approach. 

This led to the identification of several high breakdown polymers which are known to be synthesized elsewhere (for other applications) and the discovery of a larger number of novel, but easily synthesizable, polymers. One such novel polymer has already been synthesized and is undergoing further processing for testing and validation. Plans are underway to synthesize and test more of the suggested polymers. I hope the outcomes of my research will directly contribute to the research and discovery of next-generation polymer dielectrics for high-energy, high-power applications.