Machine Learning–Based Prediction and Comparison of Numerical and Theoretical Elastic Moduli in Plant Fiber–Based Unidirectional Composite Representative Volume Elements

Natural fiber-reinforced unidirectional composites are increasingly adopted in modern industries due to their superior mechanical performance and desirable properties from both material and engineering perspectives. Among various approaches, representative volume element (RVE) generation and analysis is considered one of the most suitable and convenient methods for predicting the elastic moduli of composites. The main aim of this study is to investigate and compare the elastic moduli of natural fiber–reinforced unidirectional composite RVEs using theoretical, numerical, and machine learning models. The numerical predictions in this study were generated using the ANSYS Material Designer tool (version ANSYS 19). A comparison was made between experimental results reported in the literature and different theoretical models, showing high accuracy in validating these numerical outcomes. A dataset comprising 1600 samples was generated from numerical models in combination with the well-known theory of RVE, namely rule of mixture (ROM), to train and test two machine learning algorithms: Random Forest and Linear Regression, with the goal of predicting three major elastic moduli—longitudinal Young’s modulus (E11), in-plane shear modulus (G12), and major Poisson’s ratio (V12). To evaluate model performance, mean squared error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R2) were calculated and compared against datasets with and without the theoretical values as input variables. The performance metrics revealed that with the theoretical values, both Linear Regression and Random Forest predict E11, G12, and V12 well, with a maximum MSE of 0.033 for G12 and an R2 score of 0.99 for all cases, suggesting they can predict the mechanical properties with excellent accuracy. However, the Linear Regression model performs poorly when theoretical values are not included in the dataset, while Random Forest is consistent in accuracy with and without theoretical values.