Machine Learning-Based Analysis for Thermal Properties of MWCNTs/Polymer Composites

Abstract

This article discusses thermal prediction and data-based prediction of E-glass fiber polymer composites reinforced with multi-walled carbon nanotube (MWCNT). The hand lay-up technique was used to make the laminates by the use of woven roving E-glass mats oriented at 0º/90º, 0º/45, and 0º/135º. The volume fractions of MWCNTs into the polymer matrix were 0%, 1%, 3%, 5% and 7%, which developed fifteen different composite configurations. The Thermogravimetric Analysis (TGA) was used to find out the thermal properties of the material including weight loss, degradation and decomposition behavior. These findings indicate that the orientation of fibers and MWCNT content are crucial determinants of thermal stability and that increased resistance exists at maximized filler loadings. A number of machine learning models were developed and trained using experimental data to increase predictive power. The performance of the models was measured using standard evaluation metrics and the importance of features analysis was conducted to identify important parameters on the thermal behavior. The integrated experimental and machine learning method offers a developed framework to predict thermal properties and optimize composite design, valuable information to be used in the advanced structural and thermal applications.