Machine learning and FEA-based optimization of reinforced concrete strength and durability

Abstract

This research is groundbreaking in its combination of machine learning and finite-element modeling to assess M30-grade concrete mixtures, which include 53-grade Ordinary Portland cement, ground granulated blast-furnace slag, and basalt fiber, all at a water-to-cement ratio of 0.35. Sixteen different mix designs were evaluated for their compressive strength and corrosion characteristics. Tests on 150 mm cubes revealed that Sample 10 was the best, reaching a compressive strength of 36.5 MPa after 28 days with a displacement of 0.013 mm. Corrosion was measured in a 3.5% NaCl solution using a four-electrode macrocell setup, with simulations conducted via COMSOL Multiphysics. Machine learning models such as random forest (RF), extreme gradient boosting (XGBoost), and support vector regression (SVR) were employed to predict compressive strength and corrosion metrics. RF demonstrated the highest accuracy 0.401–0.704 V, 4.50 × 10⁻⁷-1.65 × 10⁻⁵ A cm^-2). XGBoost (MAE: 0.4–0.5, R²: 0.90) and SVR (MAE: 0.55–0.7, R²: 0.83) showed moderate and lower accuracy, respectively. This integrated RF-FEM approach offers high predictive accuracy. It also presents a novel framework that combines mechanical and corrosion modeling in SCM-modified concrete.