Optimizing transparent fiber-reinforced polymer composites: A machine learning approach to material selection

Transparent glass fiber-reinforced polymers (tGFRP) are novel high-performance materials combining optical transparency with the high mechanical properties of composite materials. Achieving high transparency requires finding the optimal combination of different glass fibers and transparent polymers with similar refractive indices from a wide range of available materials. Since it is time-consuming to manufacture numerous samples, existing analytical modeling approaches can be used to predict the best possible material combinations. To enhance the predictive capability a newly developed machine-learning model for tGFRP can be used. The developed model enables the prediction of the light transmittance of such transparent composites using an established analytical approach enhanced by data-driven machine-learning (ML) techniques. A dataset comprising two types of glass fibers and eleven different polymer compositions was compiled. An analytical model and a classical multilayer perception (MLP) serve as baseline predictions to compare the improved performance of the developed machine learning architectures that incorporate physics-informed equations and Long Short-Term Memory (LSTM) layers. The presented machine learning models can accurately predict visible light transmittance, with the best-performing model achieving an acceptable mean squared error. This predictive capability can aid with the identification of the best material combination from a database of known fiber and polymer types. This optimization is expected to enhance optical properties, thereby advancing the development of next-generation transparent composite materials. • Model to predict suitable fiber–matrix pairs for transparent composites. • Methods to measure refractive index of fibers and polymers reproducibly. • Analytical vs ML models to predict transmission of transparent composites.