Using synthetic data to develop machine learning models to predict the performance of fiber- reinforced concrete

Concrete is a widely used construction material due to its high compressive strength. However, its durability is often compromised by the development of cracks caused by tensile stress within structures. These cracks can occur during the drying process, leading to water infiltration and corrosion of the concrete reinforcement, which subsequently require repair. As a result, innovative technologies, such as self-repairing concrete and crack control, have become crucial in reducing the costs associated with structural repairs. Given this context, this study investigated novel crack control technologies in concrete structures using a machine learning model that can accurately predict the performance of a specific fiber in fiber-reinforced concrete using a comprehensive dataset. The dataset was compiled from 18 studies and further augmented using synthetic data generation techniques. It encompassed 13 different fiber types and a total of 1953 fiber-reinforced concrete formulations. The computational model was then implemented in Python, and multiple linear regression (MLR), support vector regression (SVR), Random Forest and GradientBootsting techniques were employed to develop the prediction model. The results showed that Random Forest (R² = 0.887 and RMSE = 0.110), GradientBoosting (R² = 0.868 and RMSE = 0.368) and SVR models (R² = 0.856 and RMSE = 0.376) outperformed its MLR counterpart (R² = 0.587 and RMSE = 0.637). Moreover, Random Forest shows a lower RMSE, making it more suitable to accurately predict the performance of the performance of fiber-reinforced concrete.