Use of machine learning models to predict the water penetration depth in concrete
Abdulkader El‐Mir, Samer El-Zahab, Dana Nasr, Nabil Semaan, Joseph J. Assaad, Hilal El-Hassan
Abstract
Machine learning (ML) is a robust tool within the artificial intelligence domain that offers unique solutions for predictive modeling. Prediction of water penetration depth (W pen ) is crucial for assessing the durability and service life of concrete while reducing reliance on complex and time-consuming laboratory tests. This study investigates the impact of concrete composition, age, and compressive strength on W pen using a dataset of 311 concrete specimens . Multiple supervised ML models were employed in predicting W pen , including linear regression (LR), Gaussian process regression (GPR), support vector machine (SVM), random forest (RF), regression tree (RT), and hybrid RF-SVM models. Results revealed that hybrid RF-SVM model and regression tree accurately predicted W pen . The models’ performance improved by including concrete age and compressive strength . The models were validated using data from relevant literature. This research provides valuable insights into predicting water penetration depth in concrete, offers practical tools for assessing concrete durability , and offers a more sustainable approach than laboratory testing.