Micromechanical modelling and machine learning approaches for predicting effective properties of composite materials

Accurately predicting the effective properties of composite materials is critical for their design and optimisation in advanced applications. This study integrates micromechanical modelling and machine learning to develop novel efficient predictive frameworks for composite properties. Using Representative Volume Elements (RVEs) with periodic boundary conditions, a dataset of effective moduli was generated through finite element simulations. Machine learning models, including Random Forest, Gradient Boosting, and Support Vector Regressor (SVR), were tested, with Random Forest performing best in capturing complex material behaviours. A surrogate model coupled with active learning was developed for the first time in this method to enhance computational efficiency, iteratively refining predictions by targeting regions of high uncertainty. Validation against theoretical models and experimental data highlights the superior accuracy of the proposed framework, particularly in predicting transverse and shear properties. This research established a robust methodology combining micromechanics and machine learning, paving the way for advanced composite material design with reduced computational cost and enhanced predictive precision.