The review examines ML prediction of mechanical and tribological properties in fiber-reinforced epoxy, polyester and vinyl-ester composites with silica, graphene, CNT and nanoclay fillers. Studies from 2020-2025 use ANN, SVM, random forests and Gaussian processes for wear, tensile and flexural strength and stress-strain behavior. ML may reduce experiments and aid optimization, but datasets are small, features and units vary and models often fail across systems. Multi-fidelity and physics-informed ML, open databases and standards are proposed.
Key findings
- ML can predict strength and wear within specific datasets; nanofillers change performance and feature spaces; small data, feature selection and generalization dominate limitations; physics-informed, multi-fidelity and open data are priorities.
Why this matters globally
Digital materials development may reduce trial-and-error time, cost and waste, but shared provenance and uncertainty are needed to prevent computational optima from failing in manufacturing.
Thai researcher contribution
Hemaraju Raju of Rajamangala University of Technology Isan contributes to the synthesis of ML and composite engineering design.
Limitations to consider
The review may not be systematic. Reported accuracy may arise from random splits of similar formulations, leakage and no external batches. Studies with different units and validation are not comparable, and models may omit process variability, aging and manufacturability.