Title:
Predicting Ultrasonic Pulse Velocity for Concrete Health Monitoring (Prepublished)
Author(s):
Mohammad Rahmati and Vahab Toufigh
Publication:
Materials Journal
Volume:
Issue:
Appears on pages(s):
Keywords:
concrete health monitoring; extreme gradient boosting (XGBoost); feature importance; machine learning; support vector regression (SVR); ultrasonic pulse velocity (UPV)
DOI:
10.14359/51747869
Date:
6/11/2025
Abstract:
This study employs machine learning (ML) to predict ultrasonic pulse velocity (UPV) based on the mix composition and curing conditions of concrete. A dataset was compiled using 1495 experimental tests. Extreme Gradient Boosting (XGBoost) and Support Vector Regression (SVR) were applied to predict UPV in both direct and surface transmissions. The Monte Carlo approach was used to assess model performance under input fluctuations. Feature importance analyses, including the Shapley Additive Explanation (SHAP), were conducted to evaluate the influence of input variables on wave propagation velocity in concrete. Based on the results, XGBoost outperformed SVR in predicting both direct and surface UPV. The accuracy of the XGBoost model was reflected in average R² values of 0.8724 and 0.9088 for direct and surface UPV, respectively. For the SVR algorithm, R² values were 0.8362 and 0.8465 for direct and surface UPV, respectively. In contrast, linear regression exhibited poor performance, with average R² values of 0.6856 and 0.6801 for direct and surface UPV. Among the input features, curing pressure had the greatest impact on UPV, followed by cement content. Water content and concrete age also demonstrated high importance. In contrast, sulfite in fine aggregates and the type of coarse aggregates were the least influential variables. Overall, the findings indicate that ML approaches can reliably predict UPV in healthy concrete, offering a useful step toward more precise health monitoring through the detection of UPV deviations caused by potential damage.