Accelerated Discovery of Polyetherimide based Dielectric Polymers using Graph Attention Neural Networks
POSTER
Abstract
Polyetherimide (PEI) family of polymers popularly referred as ULTEM polymers are an important class of polymers which exhibit excellent thermal and mechanical properties and are widely used in producing high temperature films for capacitor applications. The properties of PEI can be greatly enhanced by modifying its backbone with various chemical, physical and structural defects along with various endcap substitutions. The large number of defects and substitutions generate a combinatorial large design space, which is difficult to explore using
known methods. Therefore, we propose a deep learning-based graph network with inbuild attention mechanism that can identify polymer systems exhibiting desired dielectric properties. The attention score of the trained model further provides crucial insights into the effect of various defects and substitutions on ULTEM. This screening process can greatly aid design of new dielectric polymer materials which are difficult to screen using conventional techniques.
known methods. Therefore, we propose a deep learning-based graph network with inbuild attention mechanism that can identify polymer systems exhibiting desired dielectric properties. The attention score of the trained model further provides crucial insights into the effect of various defects and substitutions on ULTEM. This screening process can greatly aid design of new dielectric polymer materials which are difficult to screen using conventional techniques.
*This work was supported by the ONR through a Multi-University Research Initiative (MURI) under grant number (N00014-17-1-2656). The simulations were performed at the ALCF under the DOE INCITE and Aurora Early Science programs and at the CARC, USC.
Presenters
-
Ankit Mishra
- Univ of Southern California