Microstructural, mechanical characterization and ANN prediction of concrete with glass fiber roving waste

This study presents an experimental and computational framework for the production of sustainable concrete reinforced with glass fiber roving waste (GFRW). This study combines laboratory testing, microstructural analysis, and artificial neural network (ANN) modeling to assess and enhance the mechanical properties of GFRW-reinforced concrete. A total of 145 samples were cast using different mix proportions and fiber conditions, with fiber contents of 1%, 2%, and 3% and lengths of 3, 6, 9, and 12 mm. The mixtures containing 6 mm fibers at about 1-2% volume fraction showed the most balanced performance, producing clear gains in compressive, tensile, and flexural strength compared with the control mix. The ANN model-trained with five-fold cross-validation-achieved strong predictive accuracy (R² > 0.93, RMSE < 2.0 MPa) and successfully captured the nonlinear relationships between mix variables. Microstructural observations from FE-SEM and EDX analyses confirmed the improved fiber-matrix bonding at these optimal fiber settings. Together, the findings show that combining machine learning with experimental and microstructural evaluation can lead to data-driven mix optimization while supporting the reuse of industrial glass fiber waste in sustainable concrete development.