On the flammability and mechanical responses of abaca-reinforced composites: a machine learning-oriented approach

This paper presents a comprehensive investigation into the predictive modeling of eco-friendly composite materials reinforced by Abaca fibers, Halloysite Nanotubes (HNT), and Egg Shell Powder (ESP) additives. Deep Neural Networks (DNNs) were employed to capture the complex behaviors of these materials under various testing conditions, including Cone Calorimeter Tests (CCT), micro-indentation with Vickers and Conical indenters, and three-point bending tests. The trained DNNs exhibited remarkable accuracy in predicting critical parameters such as Heat Release Rate (HRR), Average Rate of Heat Emission (ARHE), Total Heat Release (THR), Total Smoke Production (TSP), and Total Oxygen Consumption (TOC) during CCT. Additionally, the DNNs successfully replicated force-depth diagrams from Vickers and Conical indentations, showcasing their proficiency in modeling loading and unloading profiles. Furthermore, the flexural responses during three-point bending tests were accurately predicted, encompassing flexural modulus, the maximum flexural stress, strain at break, and energy. Validation through metrics such as Mean Squared Error (MSE) and coefficient of determination (R2) demonstrated the reliability of the DNNs in capturing material behaviors. Consequently, the study showcases the potential of machine learning, particularly DNNs, as a robust tool for predictive modeling in material science so that R2 values below 0.995 were not obtained for the trained DNNs as well as MSEs in the order of e−4.