Experimental characterization and machine learning-based prediction of the thermo-viscoelastic behaviour of graphene nanoplatelet-modified GFRP laminates

This study aims to characterise the combined influence of fibre orientation and graphene nanoplatelet (GNP) modification on the thermo-viscoelastic response of glass fibre-reinforced polymer (GFRP) composites and to establish a machine learning framework to predict this response across all tested configurations. Dynamic mechanical analysis (DMA) was performed in three-point bending mode on autoclave-moulded neat and 0.25 wt.% GNP-reinforced GFRP laminates with 0° and 90° fibre orientations, over 30-120°C at four frequencies (1, 10, 25, and 50 Hz). The experimental dataset comprised 4500 data points. Time-temperature superposition (TTS) master curves were constructed at reference temperatures of 83.9°C (neat GFRP) and 97.7°C (GNP-GFRP), fitted using a generalised Maxwell model with N = 8 elements (R ²  > 0.99). Four machine learning models were trained on an 80/20 split and validated by 5-fold cross-validation. The random forest (RF) model achieved R ² = 0.9996 ± 0.0001 for storage modulus, R ² = 0.9915 ± 0.0023 for loss modulus, and R ² = 0.9964 ± 0.0009 for tan δ, substantially outperforming linear regression, SVR, and MLP-ANN. GNP addition elevates T _g by approximately 10°C and increases the equilibrium modulus from 9700 to 11200 MPa, demonstrating effective thermal stiffening of the GFRP composite.