Hybrid experimental and machine learning approach for predicting mechanical properties of carbon-palm fiber composites

Abstract

The Sustainable production of composite materials, like carbon-palm fiber composites, has been of interest due to its eco-friendliness and applicability in high-performance uses. In this article, it is proposed to use a hybrid experiment-based and machine learning approach using Recurrent Neural Networks (RNN) to predict the mechanical properties like Tensile strength and wear test of carbon-palm fiber composites from palm biomass. Experiments results for tensile and wear tests on composites with various carbon to palm fiber ratio were collected following standard test protocols for the testing of biomass-based materials. The database contained important parameters such as fiber loading, matrix, and process parameters. These were used for training and cross-validation of a trained RNN model that can model the complex, non-linear behavior between composite composition and mechanical response. The RNN architecture was optimized using cross-validation and hyper parameter optimization to yield reliable predictive accuracy. Predictions showed that the RNN model can predict tensile strength, compressive strength, and flexural modulus average absolute error less than 5% relative to experimental values. The model can also identify driving parameters on mechanical properties, i.e., carbon content and fiber orientation, and provide useful data for optimizing composites. The cooperation increases material design efficiency through the reduction in needing large experimental trials, presenting a green method for fabricating high-performance composites. The union of machine learning and experimental data provides avenues for scalable data-driven composite material design, particularly for the construction, automotive, and aerospace industries.