A Study on Machine Learning-based Property Prediction of Thermal-conductive Polymer Composites

Abstract

In response to climate change, the automotive industry has focused on developing lightweight and environmentally friendly vehicles, with active research having been conducted to enhance the energy efficiency of electric and hybrid vehicles. In this context, the development of polymer composites with superior thermal conductivity has been recognized as critical to meeting mechanical property requirements. Particularly in the early stages of research, the challenge of predicting properties based on composition ratios has become a significant concern. This thesis has presented a machine learning framework that utilized 1,774 experimental data points to predict various properties of polymer composites, such as density, heat distortion temperature, flexural modulus, flexural strength, tensile yield strength, impact strength, and thermal conductivity. The various data representation methods for composition information were employed, and the XGBoost model was trained, achieving high accuracy with an average R2 score of 0.95. Furthermore, the performance of the model was enhanced to an R2 score of 0.97 by integrating process and property information using feature selection. Also, additional experiments were conducted on data containing missing values not included in the initial training dataset. This data were restored and utilized to validate the optimized property prediction model. The models demonstrated an average mean absolute percentage error (MAPE) of less than 10%, reflecting their substantial prediction accuracy. This machine learning framework, informed by experimental data, is a valuable tool for predicting and optimizing the properties of polymer composites.