Experimental and machine learning modeling for predicting electrical conductivity property for MWCNTs based composites

Abstract

The present study aims to identify the best suitable Machine Learning (ML) model to predict the Electrical Conductivity (EC) and also to determine the most influencing feature parameter on the EC. In the view of Electromagnetic Interference (EMI) shielding materials, the EC property is an important parameter for electromagnetic waves to absorb. ML models exhibit excellent performance in predicting the EC properties of Multi-Walled Carbon Nanotubes (MWCNTs)-based polymer composites, with the aim of capturing the complex interaction between input and output. In this study, various regression-based Machine Learning (ML) models like Linear Regression (LR), Ridge Regression (RR), Lasso Regression (Lasso R), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Gradient Boost (GB) and Random Forest (RF) were employed. Using the hand lay-up method, for proposed reinforcement of filler material percentages (0%, 1%, 3%, 5%, and 7%), for different fibre (E-glass mat) orientations of [0°/45°], [0°/90°], and [0°/135°] and for coated material with Conductive Backing Layer (CBL) or without CBL, 30 specimens were fabricated. The EC properties of these composites were obtained using a Vector Network Analyzer (VNA) in the frequency range of 8.2-12.4 GHz. During the machine learning, the following parameters were considered as input feature parameters: frequency, filler percentage reinforcement, fibre orientation, and CBL-coated material. The output parameter was the EC property of the material. RF obtained best model for EC dataset with more accuracy (R 2 as 0.99), and the influencing parameters were identified as the frequency and filler percentages. This is helpful to design the composites for EMI and microwave absorption for aerospace, defence and electronic applications.