Machine Learning for Predicting the Mechanical Properties of Composite Materials: A Comparison of Ensemble Methods and Analytical Models

This study examines how machine learning (ML) models can predict the mechanical characteristics of glass fiber-reinforced polymers (GFRPs), focusing on the axial Young’s modulus (E11), Poisson’s ratio (ν12), and in-plane shear modulus (G12). Several ML algorithms, such as CatBoost, XGBoost, and Gradient Boosting ensemble methods, were compared with traditional analytical models and Digimat- simulated data. The findings showed that ML models, especially ensemble methods, achieved higher accuracy, with R2 values above 0.997 and errors typically under 3% for E11 and 5% for G12. Critical features affecting these properties—like fiber volume fraction, matrix Poisson’s ratio, and shear modulus—were identified. Sensitivity analysis revealed fiber volume fraction as the most influential for E11, while matrix properties significantly impacted ν12 and G12. The study also points out that simpler models such as k-NN and SVR cannot capture complex feature interactions. Overall, the results suggest that ML approaches are promising for predicting material properties and optimizing composite design. Future research will aim to validate these models with larger, more varied datasets and explore hybrid techniques for further enhancement.