Session: 17-15-01: Society-Wide Micro/Nano Poster Forum

Paper Number: 99220

99220 - Exploring Polymer Thermal Conductivity Using Molecular Simulations and Machine Learning 

Bulk amorphous polymers are usually regarded as thermal insulators despite that they are ubiquitous in heat transfer applications, such as thermal interface materials, heat exchangers, and thermal energy storage media. To improve thermal conductivity (TC) of bulk polymers, efforts have focused on compositing thermally conductive fillers with amorphous polymer matrices. Different fillers, such as metals, ceramics and carbon, have been intensively investigated to enhance the TC of polymer composites. However, it has been pointed out that the polymer matrix TC has a major impact on the composite TC, with higher matrix TC leading to greater enhancement in composite TC. As a result, finding amorphous polymers that intrinsically have high TC will be of great importance to their heat transfer applications.

 

Despite the progress in understanding thermal transport physics in polymers, relatively limited information is known which type of amorphous polymer is better in conducting heat, and experimentally screening a large number of natural and synthesized polymers is very challenging. Recently, machine learning approaches have been increasingly applied for identifying or designing materials with desired properties. However, relatively limited effort has been put on finding thermally conductive polymers using machine learning, mainly due to the lack of polymer TC databases with reasonable data volume. The current largest polymer database, PoLyInfo, has a very limited number of pure polymers with TC values, and among them, many have spread values from different sources which makes the data very noisy. Hence, it would be ideal if a structure-TC relation can be established for quickly screening a large number of polymers to identify promising candidates for further experimental exploration.

 

In this work, we employ high-throughput molecular dynamics simulations to generate a TC database for over 600 polymers. We then utilize machine learning to construct surrogate models for the structure-TC relationship. Using this surrogate model, we screen the PoLyInfo database and explore polymers with relatively high TC (> 0.300 W/m-K). Polymers with a wide range of TC values are selected for validation using molecular dynamics simulations. Those polymers predicted to have relatively high TC by the machine learning model are mostly found to also have high TC values from the molecular dynamics simulations. Among them, 34 polymers were found to potentially have TC > 0.400 W/m-K, with majority of them belonging to polyamide and polysulfides classes, and the rest are polyvinyls, polyoxides, polyimines, polyolefins, and polyesters. This work may provide useful guidance to the experimental exploration of high TC polymers.

Presenting Author: Hanfeng Zhang University of Notre Dame

Presenting Author Biography: Hanfeng is a Ph.D student advised by Prof. Tengfei Luo from the University of Notre Dame.

Authors:

Hanfeng Zhang University of Notre Dame
Ruimin Ma University of Notre Dame
Jiaxin Xu University of Notre Dame
Tengfei Luo University of Notre Dame